0 Applications of Fourier Domain Mode Locked ... Coherence Tomography Imaging 2O9
advertisement
Applications of Fourier Domain Mode Locked Lasers for Optical Coherence Tomography Imaging MASSACHUSETTS INS'fE by OF TECHNOLOGY Desmond Christopher Adler AUG 07 2O9 B.Sc. Electrical Engineering University of Alberta, Canada, 2001 LIBRARIES S.M. Electrical Engineering and Computer Science Massachusetts Institute of Technology, United States of America, 2004 SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCE IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN ELECTRICAL ENGINEERING AND COMPUTER SCIENCE AT THE OF TECHNOLOGY INSTITUTE MASSACHUSETTS JUNE 2009 ARCHIVES © Massachusetts Institute of Technology. All rights reserved. The author hereby grants to MIT permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole or in part in any medium now known or hereafter created. Signature of Author: Department of Electrical Engineering and Computer Science February 27, 2009 Certified by: G. Fujimoto \JJames Professor of Electrical Engineering and Computer Science Thesis Supervisor I, Accepted by: J ,j7 I Terry P. Orlando Students on Graduate Committee Chair, Department Applications of Fourier Domain Mode Locked Lasers for Optical Coherence Tomography Imaging Submitted to the Department of Electrical Engineering and Computer Science in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Electrical Engineering and Computer Science at the Massachusetts Institute of Technology by Desmond Christopher Adler ABSTRACT Optical coherence tomography (OCT) is a micrometer-resolution imaging technique that produces cross-sectional images of sample microstructure by measuring the amplitude and echo time delay of backscattered light. OCT imaging is performed using low-coherence interferometry, typically with a fiber optic Michelson interferometer. OCT imaging has recently been performed by measuring the spectrum of the interference signal in the Fourier domain. In "swept source OCT" implementations, the interference spectra are generated with a wavelength-swept laser and photodetector. Axial image lines are obtained via Fourier transformation of the spectra. Fourier domain techniques have extended OCT imaging speeds from several thousand to hundreds of thousands of axial lines per second, enabling in vivo threedimensional (3D) OCT. Development of the Fourier Domain Mode Locked (FDML) laser has significantly improved the imaging performance of swept source OCT by providing an unparalleled combination of high sweep rates, large tuning ranges, narrow instantaneous linewidths, and low phase noise. This thesis develops a number of advanced OCT imaging applications using FDML laser technology. Ultrahigh-speed sub-nanometer phase profilometry is performed by measuring the phase of the OCT interference signal, taking advantage of the inherent phase stability of FDML lasers. Extending this concept, phase-sensitive OCT is used to detect gold nanoshell contrast agents with extremely high signalto-noise ratios by inducing photothermal phase modulations in the sample. Working in collaboration with industrial partners, a 3D-OCT imaging system incorporating an FDML laser is constructed for clinical research in gastroenterology. Spiral-scanning imaging catheters are developed for use in the human esophagus and colon, enabling high-density 3D-OCT endomicroscopy of the gastrointestinal tract. Finally, clinical pilot studies are conducted in collaboration with medical partners to demonstrate the utility of 3D-OCT endomicroscopy for pathology detection, treatment planning, and follow-up assessment. The convergence of 3D spatial resolution, imaging speed, field of view, and minimally invasive access enabled by 3D-OCT are unmatched by most other biomedical imaging techniques. Though still early on in its development, 3D-OCT may have a profound impact on human healthcare and industrial inspection by enabling visualization and quantification of 3D sample microstructure in situ and in real time. Thesis Supervisor: James G. Fujimoto Professor of Electrical Engineering and Computer Science ACKNOWLEDGEMENTS I am truly blessed to have had this opportunity. The intellectual stimulation, abundant research opportunities and fertile environment found in Professor James Fujimoto's Optics and Quantum Electronics group is vanishingly rare and I am honoured to have been given the opportunity to conduct my Ph.D. work here. Of course the most valuable assets of an academic research group are the people. I owe a debt of gratitude to those who taught and mentored me, to those who worked with me in the lab and in the hospital, and to those who will continue our work long after I depart. I would like to especially thank Prof. Fujimoto for his guidance and support over the years and for his patience with a student who is still, at times, a Western Canadian hockey player in spirit. I also thank Robert Huber for his mentorship, friendship, indomitable spirit, generosity with his time, and for passing on the secrets of the Fourier Domain Mode Locked laser. I would not have accomplished much without Robert's help and I won't soon forget his contributions to my career. It was a true privilege to work as one-half of "Team FDML." To my thesis committee members, Dr. Hiroshi Mashimo and Dr. Kamran Badizadegan, heartfelt thanks for your thoughtful input and advice. Additional praise goes to Hiroshi, Laren Becker, Marisa Figueiredo, and the VA nursing staff for putting up with temperamental computers, fragile imaging probes and an endoscopy suite full of MIT engineers all for the benefit of science. I would like to thank Tony Ko, Aaron Aguirre, Pei-Lin Hsiung, and Andrew Kowalevic for paving the way before me and for showing me what a graduate student should aspire to. Thank you very much to Joe Schmitt and Bob Shearer at LightLab Imaging Inc. for supplying the imaging engine, early catheters, and advice that made much of the clinical studies possible. For the many therapeutic lunchtime conversations, I thank Vivek Srinivasan, Kenzie MacIsaac, and Alex Park. The food wasn't terrific but the company always was. To Maciej Wojtkowski and Iwona Gorczynska, thank you for your enduring humour in the face of near-constant crisis. There must be something in the water in Poland. I am grateful for the earlier work of Yu Chen and Paul Herz in endoscopic OCT and for leaving me with a smoothlyrunning collaboration at the VA Boston Healthcare System. More recently it has been a pleasure working with Chao Zhou, Tsung-Han Tsai, Hsiang-Chieh Lee, Ben Potsaid, and Shu-Wei Huang. Thank you especially for putting up with my music in the lab. The future of the group is truly in good hands with you. Jonathan Liu and Yueli Chen have also been good friends and I wish them the best of luck. And of course I thank Dorothy Fleischer for all of the administrative assistance and ever-present candy dish over the years. I must thank the Government of Canada and the Province of Alberta for the funding assistance during my studies. The Natural Sciences and Engineering Research Council scholarship and the Sir James Lougheed Award were welcome reminders of my country's support. Finally, I save my last and most appreciative thanks for my family and friends. I know that the Ph.D. process has been as challenging for my wife Annie Kim Adler as it has been for me, especially since she agreed to move 4000 km away from a happy life and warm weather in Vancouver to join me here. Still, Annie has been there every day with me to celebrate when progress was good and for a shoulder to lean on when progress was less-than-good. My parents, Desmond Alfred Adler and Eva Adler, and my little sister, Heather Adler, have, along with Annie, been a source of constant encouragement from the day I decided to embark on this adventure and my love goes out to them for it. My new family members, John, Susan, Peter and Nickey Kim, my honourary "second set of parents", David and Nelly Lloyd, and the rest of my family and friends have also been wonderfully supportive. Finally I thank Scott and Jennifer Rice, Cam and Berengere Moore, and Darren Dimarco for never letting me forget that, no matter where life takes me, I'll always be a prairie boy. My hope as a departing graduate student is that I have left the group in a better position than when I began. If this is true, it is only because of the people around me. Before the Snow Now soon, ah, very soon, I know The trumpets of the north will blow, And the great winds will come to bring The pale, wild riders of the snow. Darkening the sun with level flight, At arrowy speed, they will alight, Unnumbered as the desert sands, To bivouac on the edge of night. Then I, within their somber ring, Shall hear a voice that seems to sing, Deep, deep within my tranquil heart, The valiant prophecy of spring. -- W. Bliss Carman 1861-1929 This thesis is dedicated to my wife, Annie, and to our home, the True North strong and free. TABLE OF CONTENTS CHAPTER 1: PRINCIPLES OF SWEPT SOURCE OPTICAL COHERENCE TOMOGRAPHY....... 8 ........................ 9 ....................... 1.1 Introduction to O C T Im aging .......................................... 9 1.2 Introduction to Swept Source Detection ...................................................... 10 1.3 Sw ept Source O CT Theory ................................................................................... 11 1.3.1 Signal to Noise R atio ........................................................................................................ 12 ................... ........................................ Speed Imaging and 1.3.2 Sensitivity 13 ........................ ...................................... 1.3.3 Axial Resolution and Imaging Depth 14 ............... 1.4 Com parison to Spectral Dom ain OCT .................................................................... 17 1.5 F igures .............................................................................. 18 ....................................................................................... 1.6 References CHAPTER 2: PRINCIPLES OF FOURIER DOMAIN MODE LOCKED LASERS ............................ 22 2.1 Conventional Wavelength-Swept Lasers ........................................................ 23 2.2 FDML Theory and Typical Performance ........................................................ 24 ............... 25 ....................... 2.2.1 Theory of FD M L Operation ................................................ 26 ................................... 2.2.2 Typical FDM L Perform ance ................................. 2.3 B uffered C avity D esigns .......................................................................................................... 27 2.3.1 Single-Stage B uffering ....................................................... ................................... 28 ................ 29 2.3.2 D ouble-B uffering ........................................... ...................................... 30 .......... 2.3.3 Future Designs and Higher Order Buffering ..................... 31 2.4 Dispersive FDML Cavities and Sigma Ring Designs ............................. ......................... 32 ...... .. .. ... ................. 2.4.1 Polarization Chrom aticity ...................... ....................... 32 2.4.2 Dispersive FDML Theory ........................................ ............. 33 ...................... 2.4.3 FD ML Lasers at 1060 nm .................................................. 2.4.4 Low Speed Broadband FDML at 1310 nm .................................................................... 35 .... .... . .... ................................................. 37 2.5 Figures ......................... ........ ............................................. 4 4 .............. ........ .................... .................................. ces 2 .6 R eferen CHAPTER 3: PHASE SENSITIVE OCT USING FDML LASERS ................................................... 46 47 .............. ................ .............. ......... 3.1 M otivation .................................. 48 .................................. Lasers FDML of Measurements Stability 3.2 Phase 48 ................ .............................................. ..... ... 3.2.1 E xperim ental Setup ........................ 49 3.2.2 D ata Processing ...................... .. ........ ....... ................................................... 52 3.2.3 Conventional Swept Laser ...................................................... ....... ...... ....... ................................. 53 3.2.4 Non-Buffered FDML Laser .......................... ..... 53 ....... ...................... .................. 3.2.5 Buffered FD M L Lasers ................................ 55 ........... 3.3 Sub-N anom eter D ynam ic Sensing .................................... .......... ........................... 55 3.4 Sub-N anom eter 3D Phase Profilom etry ........................................ .................................. 57 ...................................... 3 .5 F igures .................................................................................. 61 ....................................................... ........ ........ .. 3 .6 Referen ces ................................................... CHAPTER 4: PHOTOTHERMAL DETECTION OF GOLD NANOSHELLS WITH OCT ................ 63 . ............ ......... .............. 64 4.1 M otivation ........................................... .. .................................... 65 4.2 Sources of Contrast in O CT ...................................................... 67 ...................................... Modulation Phase Photothermal with 4.3 Imaging Gold Nanoparticles 67 ............................................. .......... ... 4.3.1 Experim ental Setup ...................................... ................................................ 68 4.3.2 Sample Preparation .......................... 68 ............... 4.3.3 D ata Processing ...................... ...... 69 4.4 Thermal Modeling ........................................................... 4.4.1 Phase to Tem perature Conversion ........................................................................ 69 71 .......................... 7............................... 4.4.2 Estim ated Therm al R esponses ........................... 72 ........................................................................................ esults R ental 4.5 Experim 74 .................................................. Frequency Modulation 4.5.1 Signal to Noise Ratio versus 75 ........................................ 4.5.2 Effect of Measurement Time on Signal to Noise Ratio ...................... 75 4.5.3 Comparison to Model Results ...................................... 76 ............................ ........................................... 4.6 Lim itations ........................................ 79 ............................................................................ ....... .. 4 .7 F igures ....................................... 83 .......................... .......................... .......... 4 .8 Referen ces ........................................ ........... CHAPTER 5: 3D-OCT PLATFORM FOR CLINICAL GASTROENTEROLOGY .......................... ................................ ..... .................................... 5.1 M otivation .................................... ......................... ........ 5.2 System Description ........................... .... ..................................... ..... ....... 5.3 Laser D esign and Optim ization ....................... 5.3.1 Sw eep Rate ................................................................. ... .... ............ ........ ...... 5.3.2 Buffered versus Non-Buffered Cavity Selection ...................................... 5.3.3 Sweep Linearization .............................................. ....... . ........................ 5.4 Optical Frequency Clock Optim ization ....................................... 5.4.1 Tradeoffs Between Imaging Depth and System Noise ............................................... 5.4.2 Tradeoffs Between Imaging Depth and Axial Resolution ............................................. 5.5 System C haracterization ................................................. .......... ....... .................................. 5.5.1 A xial Resolution .......... ........... ............ ....................................... ................................... 5.5.2 Sensitivity Rolloff ..................................... 5.5.3 True Spatial Resolution ............................................... ....................... 5.6 D ata Post-Processing ......................... ... ............. ........ ........................... .............................. 5.6.1 Fram e Flattening ......................................................................... 5.6.2 JPE G C onversion .......................... ......... ..... ................................................................. 5.7 Im age V isualization .................................................... ............................................................. ..... 5.7.1 3D Rendering ............................................................................ ... 5.7.2 Orthoplane Sectioning ........................................ ................................................ 5.7.3 Projection Viewing ........................................ 5.7.4 Linear En Face Images ..... ...................................... ......... 5.8 Figures ........................ . ........... ........................................................... 5.9 R eferen ces ............................................ 88 89 89 91 92 93 94 96 96 97 98 98 99 99 100 100 102 102 102 103 103 105 105 112 CHAPTER 6: HIGH SPEED IMAGING PROBES FOR CLINICAL 3D-OCT ................................. ........................... ....... ........... ............................ 6.1 M otivation ................................... 6.2 Fused Fiber Lens Systems ......................................... 6.2.1 ZEMAX Modeling - Standard Multimode Fiber ....................................................... 6.2.2 ZEMAX Modeling - Custom Multimode Fiber ........................................ 6.2.3 Polishing ........................ ............................................................... ................ . ..................... ....... ...... 6.2.4 Fabrication Tolerances ....................... ........................................ 6.2.5 Measured Performance ...... ................................................ 6.3 Micro-Optic Lens Systems ................................................ 6.3.1 ZEMAX Modeling ...................... ... .. ........ 6.3.2 Minimization of Backreflection ................. ....................................... 6.3.3 Measured Performance ....... 6.4 Mechanical Design .......................................................................................... 6.4.1 Torque C oil Selection ............................ ... .......... .......... ............... ....................... 6.4.2 Proximal Joint ............................................. .................... ..................... ......... ....................... ............................................. 6.4.3 Fiber Connector and Flush Port ...... 114 115 115 116 117 117 117 118 119 119 120 120 121 122 122 123 ........................................ 6.4.4 Future Probe Design ....... 6.5 Imaging Performance Comparison .......................................... 6 .6 F igures .. . ......... ........................................................................................................................... ......................................... 6.7 References ... 123 123 12 5 129 CHAPTER 7: CLINICAL 3D-OCT IN THE UPPER GASTROINTESTINAL TRACT ................... 130 . 131 ................................. ..................... 7.1 M otivation ............ 132 ........................................... .......................................................... .... 7.2 C linical Protocol .. 134 7.3 N orm al Esophagus ......................... ... ........ ................................................................... 134 ........................................ 7.3.1 Characteristic Features ...... ............................. 136 7.3.2 C om parison to H istology ................................................................ 136 ........................................ Ablation Pre-Radiofrequency Esophagus: Barrett's 7.4 137 .............................. .... ................... .. ......... 7.4.1 C haracteristic Features.......................... 138 ... ... .. ................................. 7.4.2 Comparison to Histology..................................... 139 7.5 Barrett's Esophagus: Post-Radiofrequency Ablation ............................. ..... ....................... 139 7.5.1 Characteristic Features................................................... .. ................ ............................................... 141 7.5.2 Comparison to Histology....................... ........................... 14 1 7.6 E sophageal N odules ...................... ........... ............................................. 7.6.1 C haracteristic Features.............................. . .. .............. .............................................. 142 ...... ... .... ........................ 143 7.6.2 Comparison to Histology............. ...................... ................... 143 ........................... Resection Mucosal Post-Endoscopic 7.7 Esophageal Nodules: 143 ........................ ..... ........... 7.7.1 Characteristic Features..................... .............. 144 7.7.2 Comparison to Histology.............................. 146 ...... ................ ......................................... 7.8 Figures ..................... ........ .. 157 .. .................................... .............. ......... ..... ............... ..... 7.9 R eferences ....................... CHAPTER 8: CLINICAL 3D-OCT IN THE LOWER GASTROINTESTINAL TRACT ................. . .. .. .............................................. ........................... 8.1 M otivation .. ...... .................................. ....................... Protocol 8.2 Clinical ......................... .. ..................... ...... 8.3 N orm al C olon ....................... .. ................... .. ..................... ............. ....................... 8.3.1 Characteristic Features ......................... 8.3.2 Comparison to Histology .......................................... ...................................... 8.4 Inflammatory Bowel Diseases ....... ................. 8.4.1 C haracteristic Features .......................... ..... ........................................... .... .................................. 8.4.2 Com parison to H istology ................................... ....... ...... 8.5 Radiation Proctitis: Pre-Radiofrequency Ablation ........................................ ....................................... 8.5.1 Characteristic Features ....... ........................ ........... ................................. 8.5.2 Comparison to Histology 8.6 Radiation Proctitis: Post-Radiofrequency Ablation ........................................ ........................................ 8.6.1 Characteristic Features ...... 8.6.2 Comparison to Histology ............................................ .... .......... ............................... .... .... ................ .. . .............. 8.7 Figures .......... 8.8 References ............................................. ................. ................................ 159 160 161 162 163 164 164 165 166 166 166 168 168 168 169 171 178 CHAPTER 9: CONCLUSIONS, FUTURE WORK, AND PUBLICATIONS ................................. ................. ................. ............ 9.1 Summ ary of Thesis W ork ............................. ..................................................... . .. ......... ...... 9.2 Future Work ......................... 9.3 Publications Produced During Thesis W ork ......................................................................... 179 180 181 182 CHAPTER 1 1.0 Principles of Swept Source Optical Coherence Tomography 1.1 Introduction to OCT Imaging OCT is an imaging technology that enables micron scale, cross-sectional and 3D imaging of sample microstructure in real time [1-3]. For biomedical applications, OCT can function as a type of "optical biopsy". Tissue microstructure can be imaged with resolutions approaching that of excisional biopsy and histopathology without the need to remove and process tissue specimens [4-6]. For material inspection applications, OCT can provide nondestructive 3D analysis of depth-resolved sample features. OCT is analogous to ultrasound B mode imaging, except that imaging is performed by measuring the echo time delay and intensity of back-reflected or backscattered light rather than sound. An optical beam is scanned across the sample and echoes of backscattered light are measured as a function of axial range (depth) and transverse position, as shown in Figure 1.1. Three-dimensional imaging can be conducted by performing a two-dimensional scan pattern at different transverse positions. Three-dimensional OCT (3D-OCT) enables powerful methods for visualizing tissue architecture. 3D-OCT generates comprehensive, volumetric data sets which can be used to construct arbitrary cross-sectional images, projections along arbitrary axes, or 3D renderings similar to those used in MRI or CT. Unlike confocal microscopy, 3D2 OCT enables imaging over a comparatively large field of view (up to -200 mm ) while maintaining resolutions on the order of 10 tm, and can therefore provide architectural context in addition to high- magnification views of focal abnormalities. OCT is based on a technique known as low coherence interferometry, which has been previously applied in photonic devices as well as in biological systems to perform optical ranging [7-9]. OCT measurements are performed using a Michelson interferometer with a low coherence length light source as shown in Figure 1.2. One arm of the interferometer contains a modular probe that scans the light beam over the sample and collects the backscattered light. Fiber optic catheters and endoscopes have been developed for imaging inside the body [6], and microscopes can be used for imaging excised tissue specimens or material samples. A second arm of the interferometer has a scanning reference path delay that is mechanically translated over the desired imaging depth in a classical "time domain OCT" configuration. Optical interference between the light from the sample and reference occurs only when the optical delays match to within the coherence length of the light [10, 11]. "Coherence" refers to a temporal property of the light, which is inversely related to its wavelength or frequency bandwidth. Low coherence interferometry enables the echo delay time and magnitude of backscattered light from internal tissue microstructures to be measured with high time resolution and sensitivity. 1.2 Introduction to Swept Source Detection Within the past 3 to 4 years, dramatic advances in OCT technology have resulted in 10 to 100 fold increases in imaging speed [12-14]. These advances are key to enabling 3D-OCT for in vivo biomedical applications and real-time material inspection applications. Volumetric data acquisition with 3D-OCT will enable new visualization and processing techniques such as the generation of cross-sectional images with arbitrary orientation, the generation of projection views similar to en face microscopy images, improved quantitative measurements of morphology, improved image processing techniques to reduce speckle and enhance contrast, and virtual manipulation of sample geometry for the visualization of structural morphology. These new OCT detection techniques, known as Fourier domain detection, can achieve very high speeds and sensitivities by measuring backscattered light in the Fourier domain [12-15]. Conventional OCT directly measures the interference signal, while Fourier domain OCT measures the spectrum of the interference signal. The OCT axial scan is then constructed by Fourier transformation. While this requires spectral measurement and additional signal processing steps, it has the advantage that all depth positions in the sample are measured simultaneously rather than sequentially as in conventional OCT detection techniques. Several groups working independently demonstrated in 2003 that Fourier domain detection enables 10 to 100 fold improvements in detection sensitivity, which gives corresponding improvements in the imaging speed [12-14]. Fourier domain OCT can be performed using two complementary techniques known as spectral domain OCT and swept source OCT (also known as optical frequency domain imaging or OFDI). Spectral domain detection uses a spectrometer and high-speed line scan camera to measure the spectrum of the OCT interference signal. Spectral domain OCT typically operates at 800 nm wavelengths with axial imaging rates of 29,000 - 75,000 lines per second (29 - 75 kHz) [16-18]. This technology has had a powerful impact on ophthalmic OCT imaging because it enables ultrahigh image resolutions as well as 3D-OCT imaging of retinal pathologies [19, 20]. In contrast, swept source OCT uses a wavelength-swept laser light source and a balanced pair of photodetectors to measure the interference spectrum [21-24] as shown in Figure 1.3. Swept source OCT technology has the advantage that it can perform imaging at longer wavelengths of 1000 nm and 1300 nm. Imaging at these wavelengths is important because it reduces optical scattering and improves image penetration depths [5]. 1.3 Swept Source OCT Theory Swept source OCT is similar in nature to optical frequency domain ranging (OFDR) techniques that are commonly used in telecommunications applications. In the most basic swept source OCT implementation using a single photodetector as shown in Figure 1.3, the current output of the detector is given by idet (t)r (P,+ P fr2 (z)dz+2 pr r(z)F(z)cos(2k(t)z+ B(z))dz) (1.1) where iq is the photodetector sensitivity, q is the electrical charge constant, hv is the photon energy, Pr is the reference arm power returned to the detector, and P is the power incident on the sample [23]. The sample and light source characteristics are represented by the relative axial coordinate z, the amplitude and phase of the sample reflectance profile r(z) and b(z), the instantaneous coherence function of the laser source F(z), and the time-varying wavenumber of the laser output k(t)= 2n~ / (t). At z= 0, the optical path length of the sample arm is matched to the reference arm. The first term in Equation (1.1) results in a constant DC photocurrent due to reference arm power on the photodetector. The second term results in a DC photocurrent that varies with the intensity of the backscattered sample light and is commonly referred to as the "autocorrelation term." The third term is the desired interferometric photocurrent that encodes the axial reflectance profile of the sample. The complex sample reflectance profile can be recovered by Fourier transformation of the photocurrent, which is typically carried out following digitization with a discrete Fourier transform (DFT) algorithm such as the Fast Fourier Transform (FFT). 1.3.1 Signal to Noise Ratio Swept source OCT systems provide order-of-magnitude improvements in signal to noise ratio (SNR) compared to previous time domain OCT methods. Theoretical SNR performance, as described in ref. [23], requires an understanding of the signal power F2 and the time-averaged noise power (F,2) where F and F, are the DFT values of the signal and noise photocurrents is and i,. For a single sample 2 reflector located at z = zo and having a fixed reflectivity r , and assuming that the instantaneous linewidth of the laser source is sufficiently narrow to give F (z), the signal current becomes is (t)='q 2 (1.2) IPcos(2k(t)z o ) where P = r2 p0 is the optical power reflected by the sample and returned to the photodetector. The timeaveraged noise power (i 2 (t)) is given by +2 =ith2 4 +) + RN(Pr + 2 )2 BW (1.3) The three terms in Equation (1.3) represent thermal, shot, and relative intensity noise (RIN) of the laser source in Hz -'. Here, i, is the thermal noise current in the photodetector and BW is the detector bandwidth in Hz. The SNR of a swept source OCT system is given by SNR = IF (z o ) / F,2 ) where F and F are the DFTs of the signal and noise currents, respectively. The DFT value of either photocurrent at z = z, is given by i(km)e-j2zlm/N, F(z,)= (1.4) m=0 Here i(k) is the sampled photocurrent and N s is the number of samples acquired per source wavelength sweep. For the signal component, the DFT is zero-valued at all axial positions except for z i = z0 . Using Parseval's theorem, IFs (z0 )2 F2=N i2 , the magnitude-squared of the signal DFT is given by (N 2 /2) (i2). The factor of arises due to the positive and negative frequency peaks 2 associated with the DFT of the real-valued signal photocurrent. The power of the noise DFT is given by (F 2) = N2 (i2). The SNR for swept source detection therefore becomes 1(i 2 SNR = (i2) (1.5) 2 (i,2 Equation (1.5) illustrates one fundamental advantage of swept source detection over conventional time domain OCT methods. In time domain OCT the SNR is simply the ratio (i2) / (i2), so swept source detection gives an N, /2 benefit in terms of SNR. In most swept source OCT setups a dual-balanced detector is used instead of a single photodiode, which removes excess noise from the laser source and effectively doubles the signal level, giving an additional -3 dB improvement to SNR. If the detection system is shot noise limited, with negligible thermal noise and laser RIN, and when P, <<Pr , the SNR for a swept source system can be approximated as SNR 7Ps hv (1.6) where T, = 1/ fs is the wavelength sweep duration and f, is the frequency at which wavelength sweeps are generated by the laser source. The sweep frequency is also equivalent to the rate at which axial image lines are generated, as shown in Equation (1.1). 1.3.2 Sensitivity andImaging Speed The sensitivity of an OCT system is defined as the minimum detectable sample reflectivity. The minimum detectable reflectivity results in a signal power that is equal to the noise power, or, equivalently, results in SNR = 1. Using P, = r 2 P0 and the condition SNR = 1 in Equation (1.6), the minimum detectable reflectivity rmin 2 is given by rmin 2 = hvf s ,JP (1.7) Sensitivity is typically expressed in dB as , log Sens = -10 hT (1.8) Compared to time domain OCT, swept source detection also conveys an NS /2 sensitivity benefit directly linked to the SNR benefit. This advantage can be used to detect very weak sample reflections if the system is operated at low sweep frequencies. In practice, however, the system's finite dynamic range limits the ability to simultaneously detect strong reflections and extremely weak reflections. The sensitivity advantage has found more utility in applications where high typical reflectivity levels are detected but extremely high imaging rates are required. Swept source OCT was first demonstrated at MIT in 1997, but performance was limited by available laser technologies [21, 22]. Recent advances in wavelength-swept lasers have enabled much higher speed imaging. Swept source OCT with axial imaging rates of 19 kHz was demonstrated in 2003 [25], and 115 kHz axial imaging rates were achieved in 2005 [26]. Using advanced Fourier Domain Mode Locked (FDML) laser technology, our group recently achieved record axial imaging rates of 370 kHz, -100 times faster than standard OCT [27, 28]. These order-of-magnitude increases in imaging speed have enabled many new biomedical and industrial applications of three-dimensional OCT (3D-OCT), several of which were developed in this thesis work. 1.3.3 Axial Resolution and Imaging Depth Axial resolution and maximum ranging depth are important parameters that determine the utility of a swept source OCT system. The theoretical value of axial resolution for swept source OCT is the same as in traditional time domain OCT [1], and is related to the spectral bandwidth of the laser source used in the imaging system. Theoretical axial resolution Az for a laser with a Gaussian power spectrum is given by Az Az = = 21n2 r 02 nSA/1 (1.9) (1.9) where A0 is the centre wavelength of the laser source, ns is the group refractive index of the sample medium, and AA is the full-width-at-half-maximum (FWHM) spectral bandwidth of the laser source. Typical axial resolutions for swept source OCT systems operating at 1310 nm are 7 - 14 pm [23, 29, 30], while systems operating at 1060 nm have achieved axial resolutions of 10 - 20 gm in air [31-33]. Imaging depth limitations in swept source OCT are fundamentally different than in time domain OCT. In both types of OCT, optical attenuation in the sample decreases the number of backscattered photons returning from deeper surfaces, which decreases the signal level and limits the useable imaging range or "penetration depth". Multiple scattering effects in turbid samples such as biological tissue compound this problem for both swept source and time domain OCT. The maximum sample depth that the system can interrogate, independently of optical effects in the sample, is referred to as the "imaging depth." In time domain implementations the imaging depth is determined by the maximum path differential AL that can be obtained between the reference arm and the sample arm. Aside from the practical constraints of reference arm scanner construction, there is no fundamental limit on imaging depth for time domain OCT systems. In a swept source implementation, interference fringes are acquired in the Fourier domain and images are generated via a DFT. Deeper positions in the sample are encoded in progressively higherfrequency interference fringes, so imaging depth is therefore limited by the detection and data acquisition electronics. The photodetector system must have sufficient bandwidth to detect the interference fringe frequency 2k (t) Zmax associated with the maximum imaging depth zmax . The digitizer used to sample the photodetector signal must also have sufficient bandwidth and sampling rate to avoid attenuation or aliasing of the highest frequency fringe. Maximum imaging depth is given by [23] as (1.10) = Z max 4n A where 62, is the wavelength sample spacing following digitization. Typical imaging depths for swept source OCT systems are 1 - 5 mm, depending on the imaging rate and data acquisition performance. 1.4 Comparison to Spectral Domain OCT Spectral domain OCT is similar in concept to swept source OCT, except that a non-swept laser is used as the source and a spectrometer is used in the detection arm to record the entire interference fringe spectrum at once. This is in comparison to swept source OCT, where the spectrum is recorded point-bypoint as a function of time. Spectral domain systems typically use silicon CCD detectors with high sensitivity near 800 nm, rolling off dramatically at wavelengths longer than 1000 nm. Although spectral domain OCT systems can provide nearly optimal performance in low-scattering organs such as the eye [19], these systems are not well suited for imaging scattering tissues since scattering is significantly increased at shorter wavelengths. Imaging epithelial tissues, for example, is of great interest since the majority of human cancers originate in epithelium that lines hollow organs such as the colon, esophagus, and breast ducts. In epithelial tissue, the high density of cellular organelles such as mitochondria, nuclei, and cellular membranes results in a highly scattering medium that limits the penetration depth of light at 800 nm wavelengths. Similarly, many non-biological materials also exhibit high optical scattering at short wavelengths. Swept source OCT systems can be constructed at 1000 nm [31-33], 1310 nm [23, 29, 30], and 1550 nm [34] wavelengths by using commercially-available telecommunications devices such as semiconductor optical amplifiers, singlemode fiber components, and InGaAs photodetectors. The ability to image at longer wavelengths is an important benefit for swept source OCT, although emerging InGaAs camera technology [35] may make spectral domain OCT at 1000 nm and 1310 nm possible in the near future. Swept source OCT also provides performance advantages in terms of sensitivity rolloff compared to spectral domain OCT. In both types of system the sensitivity decreases at longer sample delays. In spectral domain OCT this effect is due to the finite spectral resolution of the spectrometer used to record the interference signal. Deeper sample positions are associated with higher frequency interference fringes, as shown in Equation (1.1). Since the spectrometer must integrate the interference signal over a finite time period, random sample motion over this period causes phase jitter in the fringes that contributes to averaging or "fringe washout." The uncertainty in the fringe measurement is more severe at higher frequencies, causing a corresponding drop in sensitivity. In swept source OCT, sensitivity rolloff is caused by the finite instantaneous coherence length of the swept laser in a manner analogous to the finite spectrometer resolution. Swept lasers can be constructed with instantaneous coherence lengths of 0.06 0.2 nm [23, 29, 36] resulting in 6 dB sensitivity rolloffs at 3 - 7 mm. Spectrometer resolutions in spectral domain OCT can be comparable, but fringe washout effects result in 6 dB sensitivity rolloffs at 1 - 3 mm [12, 37, 38]. Finally, swept source OCT has so far provided faster imaging speeds than spectral domain OCT. This has been a result of rapid developments in wavelength-swept laser technology, including the FDML laser [29], along with improvements in high-speed commercial data acquisition hardware. Axial imaging rates of up to 370 kHz have been demonstrated with swept source OCT and FDML lasers [27] and the core technology remains scalable to higher speeds. Spectral domain OCT, on the other hand, is limited by the sensitivity, integration time, and readout rates of commercial line-scanning detector arrays. While recent improvements in camera technology have enabled spectral domain OCT at speeds up to 312.5 kHz, axial resolution and sensitivity rolloff performance are markedly degraded. Progress is also limited by the ability and willingness of commercial camera manufacturers to develop higher-speed products for what are initially limited research markets. Swept source OCT, on the other hand, requires comparatively fewer simultaneous commercial advances in order to scale up imaging rates. High-speed digitizer hardware is the main limitation to further increases in swept source OCT speeds, but improved data acquisition systems are desired by a wide variety of industries including aerospace and telecommunications companies. Currently, swept source OCT has proven its value for applications requiring the fastest possible imaging speeds at wavelengths of 1310 nm and 1060 nm. For applications where slower speeds are acceptable but axial resolution is of critical importance, such as retinal imaging, spectral domain OCT remains the dominant technology. 1.5 Figures Sample 1D 2D Axial (Z)Scanning Axial (Z) Scanning Transverse (X) Scanning 3D Axial (Z) Scanning XYScanning Backscattered Intensity i7. Long Coherence Length Light Figure 1.1. OCT generates cross-sectional and 3D images of tissue microstructure by measuring the echo time delay and magnitude of backscattered light. Architectural morphology can be imaged in vivo and in real time. Short Coherence Length Light uses low coherence Figure 1.2. OCT interferometry to detect the time delay and magnitude of backscattered light. C Detector output frequency - distance Figure 1.3. Swept source OCT enables a 10-100x increase in imaging speed compared to time domain OCT. A: Michelson interferometer with path length difference AL and a wavelength-swept laser source. B: Light from the sample (dotted) and reference (dashed) paths are time delayed and interfere. C: A radiofrequency beat signal proportional to AL is produced on the detector. D: The Fourier transform of the detected signal recovers the axial reflectance profile of the sample. 1.6 References [1] D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, C. A. Puliafito, and J. G. Fujimoto, "Optical Coherence Tomography," Science, vol. 254, pp. 1178-1181, Nov 22 1991. [2] J. G. Fujimoto, C. Pitris, S. A. Boppart, and M. E. Brezinski, "Optical coherence tomography: an emerging technology for biomedical imaging and optical biopsy," Neoplasia, vol. 2, pp. 9-25, Jan-Apr 2000. [3] J. G. Fujimoto, "Optical coherence tomography for ultrahigh resolution in vivo imaging," Nature Biotechnology, vol. 21, pp. 1361-1367, Nov 2003. [4] J. G. Fujimoto, M. E. Brezinski, G. J. Tearney, S. A. Boppart, B. Bouma, M. R. Hee, J. F. Southern, and E. A. Swanson, "Optical biopsy and imaging using optical coherence tomography," Nature Medicine, vol. 1, pp. 970-972, Sep 1995. [5] M. E. Brezinski, G. J. Tearney, B. E. Bouma, J. A. Izatt, M. R. Hee, E. A. Swanson, J. F. Southern, and J. G. Fujimoto, "Optical coherence tomography for optical biopsy. Properties and demonstration of vascular pathology," Circulation,vol. 93, pp. 1206-13, Mar 15 1996. [6] G. J. Tearney, M. E. Brezinski, B. E. Bouma, S. A. Boppart, C. Pitvis, J. F. Southern, and J. G. Fujimoto, "In vivo endoscopic optical biopsy with optical coherence tomography," Science, vol. 276, pp. 2037-2039, 1997/06/27 1997. [7] K. Takada, I. Yokohama, K. Chida, and J. Noda, "New measurement system for fault location in optical waveguide devices based on an interferometric technique," Applied Optics, vol. 26, pp. 1603-1608, 1987. [8] H. H. Gilgen, R. P. Novak, R. P. Salathe, W. Hodel, and P. Beaud, "Submillimeter optical reflectometry," IEEE JournalofLightwave Technology, vol. 7, pp. 1225-1233, 1989. [9] R. Youngquist, S. Carr, and D. Davies, "Optical coherence-domain reflectometry: a new optical evaluation technique," Optics Letters, vol. 12, pp. 158-160, March 1987. [10] E. A. Swanson, D. Huang, M. R. Hee, J. G. Fujimoto, C. P. Lin, and C. A. Puliafito, "High-speed optical coherence domain reflectometry," Optics Letters, vol. 17, pp. 151-153, 1992. [11] E. A. Swanson, J. A. Izatt, M. R. Hee, D. Huang, C. P. Lin, J. S. Schuman, C. A. Puliafito, and J. G. Fujimoto, "In vivo retinal imaging by optical coherence tomography," Optics Letters, vol. 18, pp. 1864-1866, 1993/11/01 1993. [12] M. A. Choma, M. V. Sarunic, C. H. Yang, and J. A. Izatt, "Sensitivity advantage of swept source and Fourier domain optical coherence tomography," Optics Express, vol. 11, pp. 2183-2189, SEP 8 2003. [13] J. F. de Boer, B. Cense, B. H. Park, M. C. Pierce, G. J. Tearney, and B. E. Bouma, "Improved signal-to-noise ratio in spectral-domain compared with time-domain optical coherence tomography," Optics Letters, vol. 28, pp. 2067-2069, Nov 1 2003. [14] R. Leitgeb, C. K. Hitzenberger, and A. F. Fercher, "Performance of Fourier domain vs. time domain optical coherence tomography," Optics Express, vol. 11, pp. 889-894, 2003/04/21 2003. [15] A. F. Fercher, C. K. Hitzenberger, G. Kamp, and S. Y. Elzaiat, "Measurement of Intraocular Distances by Backscattering Spectral Interferometry," Optics Communications, vol. 117, pp. 4348, MAY 15 1995. [16] B. Cense, N. Nassif, T. C. Chen, M. C. Pierce, S. Yun, B. H. Park, B. Bouma, G. Tearney, and J. F. de Boer, "Ultrahigh-resolution high-speed retinal imaging using spectral-domain optical coherence tomography," Optics Express, vol. 12, pp. 2435-2447, 2004. [17] N. Nassif, B. Cense, B. H. Park, S. H. Yun, T. C. Chen, B. E. Bouma, G. J. Tearney, and J. F. de Boer, "In vivo human retinal imaging by ultrahigh-speed spectral domain optical coherence tomography," Opt Lett, vol. 29, pp. 480-2, Mar 1 2004. [18] Y. Zhang, B. Cense, J. Rha, R. S. Jonnal, W. Gao, R. J. Zawadzki, J. S. Werner, S. Jones, S. Olivier, and D. T. Miller, "High-speed volumetric imaging of cone photoreceptors with adaptive optics spectral-domain optical coherence tomography," Optics Express, vol. 14, pp. 4380-4394, May 2006. [19] M. Wojtkowski, V. Srinivasan, J. G. Fujimoto, T. Ko, J. S. Schuman, A. Kowalczyk, and J. S. Duker, "Three-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence tomography," Ophthalmology, vol. 112, pp. 1734-46, Oct 2005. [20] V. J. Srinivasan, M. Wojtkowski, A. J. Witkin, J. S. Duker, T. H. Ko, M. Carvalho, J. S. Schuman, A. Kowalczyk, and J. G. Fujimoto, "High-definition and 3-dimensional imaging of macular pathologies with high-speed ultrahigh-resolution optical coherence tomography," Ophthalmology, vol. 113, pp. 2054-2065, Nov 2006. [21] S. R. Chinn, E. A. Swanson, and J. G. Fujimoto, "Optical coherence tomography using a frequency-tunable optical source," Optics Letters, vol. 22, pp. 340-342, Mar 1 1997. [22] B. Golubovic, B. E. Bouma, G. J. Tearney, and J. G. Fujimoto, "Optical frequency-domain reflectometry using rapid wavelength tuning of a Cr4+:forsterite laser," Optics Letters, vol. 22, pp. 1704-1706, Nov 15 1997. [23] S. H. Yun, G. J. Tearney, J. F. de Boer, N. Iftimia, and B. E. Bouma, "High-speed optical frequency-domain imaging," Optics Express, vol. 11, pp. 2953-2963, Nov 3 2003. [24] W. Y. Oh, S. H. Yun, B. J. Vakoc, G. J. Tearney, and B. E. Bouma, "Ultrahigh-speed optical frequency domain imaging and application to laser ablation monitoring," Applied Physics Letters, vol. 88, pp. -, Mar 6 2006. [25] S. H. Yun, G. J. Tearney, B. E.Bouma, B. H. Park, and J. F. de Boer, "High-speed spectraldomain optical coherence tomography at 1.3 mu m wavelength," Optics Express, vol. 11, pp. 3598-3604, DEC 29 2003. [26] W. Y. Oh, S. H. Yun, G. J. Tearney, and B. E.Bouma, "115 kHz tuning repetition rate ultrahighspeed wavelength-swept semiconductor laser," Optics Letters, vol. 30, pp. 3159-3161, DEC 1 2005. [27] R. Huber, D. C. Adler, and J. G. Fujimoto, "Buffered Fourier domain mode locking: unidirectional swept laser sources for optical coherence tomography imaging at 370,000 lines/s," Optics Letters, vol. 31, pp. 2975-2977, Oct 2006. [28] D. C. Adler, R. Huber, and J. G. Fujimoto, "Phase-sensitive optical coherence tomography at up to 370,000 lines per second using buffered Fourier domain mode-locked lasers," Optics Letters, vol. 32, pp. 626-628, Mar 2007. [29] R. Huber, M. Wojtkowski, and J. G. Fujimoto, "Fourier Domain Mode Locking (FDML): A new laser operating regime and applications for optical coherence tomography," Optics Express, vol. 14, pp. 3225-3237, Apr 17 2006. [30] D. C. Adler, Y. Chen, R. Huber, J. Schmitt, J. Connolly, and J. G. Fujimoto, "Three-dimensional endomicroscopy using optical coherence tomography," Nature Photonics, vol. 1, pp. 709-716, Dec 2007. [31] E.C. W. Lee, J. F. de Boer, M. Mujat, H. Lim, and S. H. Yun, "In vivo optical frequency domain imaging of human retina and choroid," Optics Express, vol. 14, pp. 4403-4411, May 15 2006. [32] R. Huber, D. C. Adler, V. J. Srinivasan, and J. G. Fujimoto, "Fourier domain mode locking at 1050 nm for ultra-high-speed optical coherence tomography of the human retina at 236,000 axial scans per second," Optics Letters, vol. 32, pp. 2049-2051, Jul 2007. [33] V. J. Srinivasan, D. C. Adler, Y. Chen, I. Gorczynska, R. Huber, J. Duker, J. S. Schuman, and J. G. Fujimoto, "Ultrahigh-speed Optical Coherence Tomography for Three-Dimensional and En Face Imaging of the Retina and Optic Nerve Head," 2008, pp. iovs.08-2127. [34] K. Asaka and K. Ohbayashi, "Dispersion matching of sample and reference arms in optical frequency domain reflectometry-optical coherence tomography using a dispersion-shifted fiber," Optics Express, vol. 15, pp. 5030-5042, Apr 2007. [35] B. Povazay, B. Hermann, A. Unterhuber, B. Hofer, H. Sattmann, F. Zeiler, J. E. Morgan, C. Falkner-Radler, C. Glittenberg, S. Blinder, and W. Drexler, "Three-dimensional optical coherence tomography at 1050 nm versus 800 nm in retinal pathologies: enhanced performance and choroidal penetration in cataract patients," Journal of Biomedical Optics, vol. 12, pp. -, Jul-Aug 2007. [36] R. Huber, K. Taira, T. H. Ko, M. Wojtkowski, V. Srinivasan, and J. G. Fujimoto, "High-speed, amplified, frequency swept laser at 20 kHz sweep rates for OCT imaging," in Conference on Lasers and Electro-Optics/Quantum Electronics and Laser Science and Photonic Applications, Systems and Technologies 2005, Baltimore, 2005, p. JThE33. [37] M. Wojtkowski, R. Leitgeb, A. Kowalczyk, T. Bajraszewski, and A. F. Fercher, "In vivo human retinal imaging by Fourier domain optical coherence tomography," Journal of Biomedical Optics, vol. 7, pp. 457-463, 2002/07/ 2002. [38] B. Potsaid, I. Gorczynska, V. J. Srinivasan, Y. L. Chen, J. Jiang, A. Cable, and J. G. Fujimoto, "Ultrahigh speed Spectral/Fourier domain OCT ophthalmic imaging at 70,000 to 312,500 axial scans per second," Optics Express, vol. 16, pp. 15149-15169, Sep 15 2008. CHAPTER 2 2.0 Principles of Fourier Domain Mode Locked Lasers 2.1 Conventional Wavelength-Swept Lasers Swept source OCT imaging places significant performance demands on the wavelength-swept laser. The imaging speed, axial resolution, and ranging depth of the OCT system are determined by the sweep rate, tuning range, and instantaneous linewidth of the laser, respectively. The phase stability of the laser also affects the flow or displacement sensitivity of phase-sensitive OCT systems. Therefore tuning ranges of > 150 nm, instantaneous linewidths of < 0.1 nm, and sweep rates of 50 - 500 kHz are simultaneously desired [1] with an average output power of > 10 mW. Conventional wavelength-swept laser sources typically consist of a broadband gain medium with a tunable optical band-pass filter inside the cavity as shown in Figure 2.1. During laser startup, lasing must build up from amplified spontaneous emission (ASE) or fluorescence background over several cavity roundtrips. As the tunable filter switches to a new centre wavelength position, lasing collapses at the previous position and must build up again at the new position. The maximum achievable sweep rate is therefore limited by the characteristic time constant for building up laser activity inside the cavity [1]. This non-stationary operation, corresponding to a temporally varying distribution of energy between the longitudinal modes of the laser cavity, has many drawbacks, including increased amplitude noise, low power, and broad instantaneous linewidth. Conventional swept lasers experience decreased performance as the sweep rate is increased. Output power begins to decrease significantly when light at a given wavelength does not make enough cavity roundtrips to fully saturate the gain medium. This has been termed the "saturation limit" [1]. The corresponding sweep frequency fat at the saturation limit for a laser with a sinusoidally-driven tunable filter is given by sat "(2.1) log (2.1) log(Gp) Af c/ atAsw L ncAlFW f PASE Here, G is the small-signal gain of the gain medium, p is the fraction of energy fed back into the cavity after each round trip, AAf is the FWHM linewidth of the tunable filter, c is the free-space speed of light, the factor 1/i corrects for the increased tuning speed (nm/s) of a sinusoidal drive compared to a linear drive, Pa,, is the saturation power of the gain medium, AFW,, is the full-width tuning range of the laser, PASE is the total ASE power of the gain medium, L is the physical cavity length, and n c is the refractive index of the cavity material. The sweep frequency at the saturation limit represents the frequency at which the filter moves over its FWHM linewidth in a time period equal to that required for saturated lasing to build up from ASE. A full derivation of this condition can be found in reference [1]. At sweep rates above fat , the laser output power rapidly declines until the point where light makes only a single roundtrip in the cavity before being blocked by the tunable filter as it shifts to a new centre wavelength. This has been termed the "single roundtrip limit" [1]. The corresponding sweep frequency fsgi at the single roundtrip limit for a laser with a sinusoidally-driven tunable filter is given by Scz gi AAFW L nc (2.2) At frequencies near fsngi, the laser output consists mainly of filtered ASE from the gain medium and output power decreases by 2-3 orders of magnitude relative to the saturation limit. The drop in power versus sweep rate for a typical swept laser constructed with a 2.4 m fiberoptic ring cavity, a fiberoptic Fabry-Perot tunable filter (FFP-TF), and a semiconductor optical amplifier (SOA) gain chip is shown in Figure 2.2 [1]. Normalized output power traces are shown both with and without extra-cavity amplification by a second SOA. The two vertical lines mark the saturation limit (left line) and the single roundtrip limit (right line). Although in this example the saturation limit occurs at -8 kHz, it is possible to increase fat, by decreasing the cavity length. At extremely short cavity lengths the spacing between cavity modes becomes wider and a more discrete stepwise tuning rather than a smooth sweep can arise, potentially generating aliasing artifacts during OCT imaging. Several research groups and commercial entities have investigated conventional swept lasers for OCT imaging. One of the more successful designs uses a fixed diffraction grating and rotating polygon mirror as the tunable filter element. Sweep rates of 15.7 kHz - 115 kHz and full-width tuning ranges of 74 - 125 nm have been demonstrated using these lasers [2-5], although performance at the higher sweep rates is significantly degraded. Conventional swept lasers are typically restricted to sweep rates of 50 - 60 kHz for most OCT imaging applications, which can be insufficient for acquiring densely sampled 3D datasets in vivo [5]. 2.2 FDML Theory and Typical Performance Fourier Domain Mode Locked (FDML) lasers are a class of wavelength-swept laser that overcomes limitations in the maximum speed rate while simultaneously providing a broad tuning range, narrow instantaneous linewidth, and high output power. FDML lasers were originally developed by Dr. Robert Huber in 2005 while he was a visiting scientist in Prof. James Fujimoto's group at MIT [6, 7]. FDML lasers generate very low-noise frequency sweeps, equivalent to train of highly chirped laser pulses [7-10]. Figure 2.3 shows a schematic representation of the FDML concept. A dispersion managed delay line is incorporated into the cavity and the filter is tuned synchronously to the cavity round-trip time (or a harmonic of the round-trip time). This results in a quasi-stationary mode of operation. Light at one wavelength propagates through the cavity and returns to the filter input in the same amount of time required for the filter to tune through one cycle and return to the same wavelength position. Therefore lasing occurs simultaneously at all wavelengths in the sweep and does not have to continuously build up from ASE or fluorescence background. In other words, an entire wavelength sweep is optically stored within the dispersion managed delay line in the laser cavity. 2.2.1 Theory ofFDML Operation FDML operation requires precise synchronization between the optical roundtrip time of light in the cavity and the tuning period of the intracavity filter [7]. This condition is given by A rf = ncL (2.3) where rf is the tuning period of the tunable filter and A is a positive integer representing the harmonic of the cavity roundtrip time. The synchronization condition can also be expressed in terms of the gating time during which the tunable filter transmits light at a given wavelength. The required gating time rg is given by (2.4) rg = ;T fdA,FW where fd is the filter drive frequency and f, = 2 fd is the effective sweep frequency for a sinusoidal drive waveform. In practice rf must be synchronized to better than one part in lx10 relative to ncL /c in order to ensure optimal operation. Due to the stringent synchronization requirements of FDML operation, chromatic dispersion in the long laser cavity can cause decreased performance by altering the optical roundtrip time for different wavelengths in the sweep. Since the tunable filter must be operated at a single drive frequency it can only be perfectly synchronized to one wavelength in the presence of dispersion, resulting in additional loss at the desynchronized wavelengths and subsequent narrowing of the tuning range. The variation in roundtrip time Ardisp due to the non-zero dispersion slope at wavelength A in the cavity is given as Ardisp = (A - 1313nm) 2 .L 0.086ps/km.nm 2 (2.5) To minimize dispersion effects, initial FDML lasers were designed to operate at centre wavelengths near 1313 nm. This is the zero-dispersion wavelength for standard Coming SMF-28 singlemode fiber, which can be used to form the FDML cavity and is also typically used in fiberoptic components such as couplers and isolators. Dispersion effects are covered more completely in Section 2.4.2 below. Under ideal operation, sequential sweeps in and FDML laser have the same phase evolution and are mutually coherent. The filter dissipates very little energy since, at any point in time, the light incident on the filter input contains only a narrow band of wavelengths that are matched to the transmission window of the filter at that moment. In the frequency domain this requires destructive interference of all longitudinal modes that are not transmitted through the narrowband filter at a given time. Thus, the phases of the longitudinal modes must be locked. Traditional mode-locked lasers have longitudinal modes that are locked together with constant phase, corresponding to the generation of a train of short pulses at a repetition rate equal to the cavity round-trip time. FDML lasers have modes that are locked together with a different phase relationship. The laser output is not a train of short pulses, but is a train of wavelength sweeps or highly chirped, very long pulses. The tunable narrowband filtering in an FDML laser is equivalent to an infinite number of narrowband amplitude modulators that are slightly out of phase. Fourier domain mode locking is performed by periodic spectral modulation, rather than amplitude modulation. This can be viewed as the Fourier domain analog of mode locking for short-pulse generation. 2.2.2 Typical FDML Performance Figure 2.4 shows a schematic diagram of an FDML laser. The laser uses a fiber ring geometry with an SOA as the gain medium and an FFP-TF as the tunable optical bandpass filter. The cavity is formed from a 7 km length of SMF-28 singlemode fiber giving an optical roundtrip time of 34 pts, an FFP-TF drive frequency of 29 kHz, and an effective sweep rate of 58 kHz. Depending on the birefringence properties of the fiber ring, the SOA can be polarized (low cavity birefringence) or polarization-insensitive (high cavity birefringence). Polarized SOAs generally provide higher gain, broader amplification bandwidths, and increased output power compared to polarization-insensitive SOAs. The FFP-TF is selected to have a free spectral range slightly greater than the desired tuning range. The FFP-TF is driven with a sinusoidal waveform created by a high-precision digital function generator. The waveform is amplified by an electronic power amplifier for driving the low-impedance capacitive load of the lead zirconate titanate (PZT) FFP-TF actuator. The optical isolators (ISO) eliminate extraneous intracavity reflections and ensure unidirectional lasing of the ring cavity. A fiber splitter acts as the output coupler with the coupling ratio (typically 20% - 45%) controlling the tradeoff between output power and tuning range. For OCT imaging the laser output can be further amplified with a second SOA. The inset in Figure 2.4 illustrates the alternating series of long-to-short (backward) and short-to-long (forward) wavelength sweeps created by the sinusoidal FFP-TF drive. Figure 2.5 shows the transient intensity profiles of the 7 km FDML laser operating at different harmonics of the cavity roundtrip time [7] as measured by Dr. Robert Huber during initial development of the FDML concept. Average output power was 19 mW and the total tuning range was 105 nm. The data is shown for the direct laser output without booster amplification in order to ensure that the transient intensity profiles are not obscured or shaped by saturation effects [1]. In contrast to other high-speed swept lasers, the forward and backward sweeps of the FDML laser have the same intensity profile and the same maximum power for a wide range of sweep rates. Since FDML lasers operate in a quasi-stationary regime, they are inherently less noisy than conventional swept lasers. This is illustrated in Figure 2.6. Figure 2.6 (top) shows the RF interference fringes produced by an asymmetric Michelson interferometer when an FDML and a standard swept laser are used as the light source. The mutually-coherent sweeps in the FDML laser provide superior phase stability, which is discussed in more detail in Chapter 3. Due to the narrower instantaneous linewidth caused by multi-passing of the FFP-TF in the FDML laser, the OCT point spread functions shown in Figure 2.6 (bottom) roll off more slowly than when the conventional swept laser is used. Imaging depths of up to 7 mm are possible using FDML lasers, whereas a considerable drop in sensitivity is observed over only 3 mm with the conventional swept laser. From the roll off of the point spread functions, a linewidth of 0.06 nm can be calculated for the FDML laser. This is much narrower than the filter bandwidth of 0.25 nm. Broader spectral filters can therefore be applied than in conventional swept lasers, reducing component costs and losses in the cavity. 2.3 Buffered Cavity Designs For OCT imaging in highly-scattering biological tissue, detection sensitivities of better than -100 dB are typically required to ensure reasonable image quality. With a maximum incident power on the sample of 20 mW, typical detector efficiency of 50%, and accounting for typical sample arm losses of 3 dB (single-pass), Equation 1.1 indicates that sweep rates of I - 2 MHz are possible before OCT sensitivity becomes prohibitively low. FDML lasers are an ideal technology to advance OCT imaging rates towards the MHz range since they maintain performance levels as sweep rate is scaled up. In order to increase OCT imaging rates using FDML lasers the sweep speed in nm/s must be increased. The most straightforward way to accomplish this is to simply decrease the cavity length and drive the FFP-TF with a higher-frequency sine wave. This approach has several drawbacks. First, the FFP-TFs commonly used in FDML lasers exhibit strong mechanical resonances. The voltage required to tune the FFP-TF over one free spectral range (FSR) increases dramatically at sweep rates above 200 kHz and results in sweep instability. Second, despite the fact that both forward and backward sweep directions have approximately the same transient power characteristics, as the effective sweep rate is increased beyond 300 kHz the two sweep directions begin to exhibit different noise characteristics. The forward sweep exhibits a higher noise floor, decreasing the dynamic range for OCT imaging. Finally, it is desirable to produce unidirectional sweeps in order to reduce signal processing requirements for OCT. With bidirectional sweeping, for example, every second interference fringe must be reversed prior to Fourier transformation and image formation. A cavity architecture called "buffered FDML" addresses these issues, enabling ultrahigh-speed sweeping and unidirectional operation. 2.3.1 Single-Stage Buffering Buffered FDML lasers use a cavity design that optically replicates the low-noise backward sweep and removes the undesired forward sweep by using a combination of time multiplexing and gain modulation. This concept is illustrated in Figure 2.7. The interference fringes produced by an asymmetric Mach-Zehnder interferometer are shown with the desired backward sweep circled in green and the undesired forward sweep crossed out in red. A buffered FDML cavity schematic is shown in Figure 2.8. Two output couplers are placed at evenly-spaced locations within the cavity. Each output coupler extracts an optical copy of the propagating sweep, with the second coupler extracting a copy that is time-delayed by exactly one half of the cavity round-trip time. During the time normally occupied by the forward sweep, the intracavity SOA is modulated off using direct current modulation. The two copies of the remaining backward sweep are combined outside of the cavity using a 50/50 fiber splitter and boosted by an external SOA. For the configuration shown in Figure 2.8, the sweep is copied once and the layout is referred to as "single-stage buffering." The result is a series of unidirectional, low-noise wavelength sweeps generated at twice the FFP-TF drive frequency as shown in the inset of Figure 2.8. The single-stage buffered FDML laser shown here was developed in collaboration with Dr. Robert Huber as part of this thesis project [9]. The cavity length was 1.1 km, broken into two 550 m sections. This resulted in a roundtrip time of 5.4 ps and an FFP-TF sweep rate of 185 kHz. Unidirectional wavelength sweeps were generated at a record 370 kHz. This record has since been surpassed by stretched-pulse laser sources operating in the MHz range [11, 12] but high excess noise levels make these sources generally unsuitable for imaging in biological tissue. The average output power was 36 mW and the total tuning range was 100 nm with an instantaneous linewidth of 0.1 nm, giving a peak sensitivity of -98 dB and a rolloff of 12 dB over a 4 mm ranging depth in air. The shot noise limited sensitivity was 109 dB assuming a detector efficiency of 50%, sweep duty cycle of 90%, and 10 mW of average power on the sample. A comparison of dynamic range versus imaging depth for a buffered and non-buffered configuration at effective sweep rates of 370 kHz is shown in Figure 2.9. The non-buffered FDML configuration displays a clear degradation in noise performance of the forward wavelength sweep at large depths. With a sensitivity of -98 dB and typical peak backreflection levels of -50 dB in biological tissue, the desired dynamic range for most OCT applications is -50 dB. The non-buffered dynamic range falls below this level at an imaging depth of -500 tpm for the forward sweep. The buffered laser, which produces only backward sweeps, maintains > 50 dB dynamic range until 2000 pm imaging depths. The buffered FDML laser was used to demonstrate OCT imaging at axial line rates of 370 kHz as shown in Figure 2.10. 2.3.2 Double-Buffering Significant degradation in laser stability is observed when the FFP-TF is driven near a mechanical resonance frequency, making it challenging to exceed effective sweep rates of several hundred kHz simply by increasing the FFP-TF drive frequency. To overcome this limitation and enable stable, highspeed tuning, the FDML buffering concept can be extended to further multiply the sweep rate. Additional optical copies are created outside of the cavity, time-delayed appropriately, and recombined. The FFP-TF is driven near its main mechanical resonance. For the filters used in this thesis work, the main resonant peak is near 58 kHz. Complete characterization of the FFP-TF frequency response is discussed in detail in Chapter 5. The concept of double-buffering is shown in Figure 2.11. To increase the sweep speed (nm/s) while maintaining a constant FFP-TF drive frequency (Hz), the filter is overdriven with a voltage amplitude larger than necessary to span the amplification bandwidth of the intracavity SOA. In Figure 2.11 the SOA gain bandwidth is represented by the region between A and 2. The FFP-TF drive amplitude is chosen such that the useable wavelength range is scanned in a time equal to V2 of the normal unidirectional sweep duration ts , where the filter drive frequency fd = 1/ 2t,. The intracavity SOA is modulated off during the times when the filter is outside of the range A -+ A2 . Optical copies of the sweep are generated, timedelayed, and recombined to create a near-100% duty cycle train of unidirectional wavelength sweeps at four times the FFP-TF drive frequency. Double buffering has the added benefit of decreasing the portion of the sine wave used to generate the sweep, which improves the linearity of the sweep in optical frequency (k) space. Sweep copies are generated and reconstructed using a laser design constructed for this thesis work as shown in Figure 2.12. The cavity length is 3.4 km and is broken into two sections of 1.7 km each, giving a roundtrip time of 16.7 [ts and an FFP-TF drive frequency of 60 kHz. The FFP is driven with a -2x amplitude overdrive as shown in Figure 2.11. Since the FSR of the FFP-TF is only 165 nm, the overdrive results in "wrapping" of the filter transmission window back to wavelengths containing SOA gain at the edges of the FFP-TF drive wave. This results in the creation of partial wavelength sweeps bracketing the desired sweep. These partial sweeps are removed in the same manner as the undesired forward sweep by modulating off the SOA. This effect, however, can be used in the future to create higher-order buffered FDML lasers without the need for additional fiber delays or couplers as described in Section 2.3.3. After the intra-cavity SOA modulation is activated, two time-shifted copies of the backward sweep are extracted from the cavity using 80%/20% and 70%/30% fiberoptic couplers. The copies extracted from the cavity are approximately 4.0 ts in duration and are time-shifted by 8.34 ts with respect to one another. The copies are then routed to an external buffering stage consisting of an unbalanced MachZehnder interferometer with a 0.85 km path imbalance that provides a 4.17 [ts time delay. In the external stage, two additional copies of the cavity output are created. Two are time-shifted by 4.17 [is and all four copies are recombined in a final fiberoptic coupler. An oscilloscope screen capture of the laser output is shown in Figure 2.13. The top trace shows interference fringes from an asymmetric Mach-Zehnder interferometer, the middle trace shows the FFP-TF drive wave, and the bottom trace shows the intracavity SOA modulation signal. Four nearly identical sweep copies are produced per FFP-TF drive cycle, quadruples the effective sweep rate compared to the FFP-TF drive frequency. Although a double-buffered FDML laser could also be constructed by breaking the cavity into 4 equal segments, the use of an external buffering stage minimizes power losses during recombination of the sweep copies. The double-buffered FDML laser has provided good performance for a number of OCT imaging experiments [13]. Initial results gave an average output power of 62 mW with a total tuning range of 158 nm and a FWHM tuning range of 117 nm, supporting an OCT axial resolution of 8.3 [tm in air. The sensitivity rolloff was -5.5 dB at an imaging depth of 2 mm in air. The time-averaged spectrum of this laser is shown in Figure 2.15 and the OCT point spread function rolloff is shown in Figure 2.14. Acquisition of more advanced highly-polarized SOA chips has recently boosted the output power of this design to 80 mW and has improved total tuning range to > 180 nm, supporting axial resolutions of 7.3 [tm in air or -5 jtm in tissue. Peak sensitivities have reached -104 dB with 29 mW of power incident on the sample compared to a theoretical value of -111 dB after accounting for losses in the microscope sample arm of -5 dB. Versions of this laser have been constructed for use by collaborators at the University of Illinois at Urbana-Champagne as well as collaborators at the University of Washington. 2.3.3 Future Designs and Higher Order Buffering To reach FDML sweep rates of 1 - 2 MHz the buffering concept can be extended by adding additional external buffering stages. Each stage would be composed of an unbalanced Mach-Zehnder interferometer with a fiber delay line to provide an appropriate time shift. Each additional stage can further multiply the sweep rate without adding additional loss beyond the excess loss of the splitter and propagation loss of the fiber. Preliminary tests indicate that the FFP-TF can tolerate at least an additional factor of 2 increase in drive amplitude without short-term loss of stability, although no long-term testing has been conducted. There is therefore reasonable confidence that FDML lasers with sweep rates of 500 kHz, with similar tuning range and output power performance to the laser described above, can be readily constructed. This is also a promising route towards breaking the 1 MHz barrier although filter ageing may be accelerated. Since the FFP-TF is a periodic filter, overdriving the device results in "wrapping" of the transmission window when the end of one FSR is reached. Current double-buffered designs have configured the FFPTF drive wave such that the filter tuning range is centered in the middle of the SOA gain band. Overdriving the FFP-TF in the absence of SOA modulation gives two partial sweeps before and after the desired central sweep due to transmission window wrapping as the filter tunes beyond its FSR. These partial sweeps have been discarded by modulation of the SOA in the same manner that the undesired forward sweep is removed. Future buffered designs could make use of the periodicity of the filter to enable higher-order buffering without the addition of external buffering stages. This concept is illustrated in Figure 2.16. Using the same FFP-TF drive amplitude as current double-buffered designs but with a different DC voltage offset to center the sweep between two consecutive filter FSR bands, two backward sweeps can be generated from 1/2 of the FFP-TF drive cycle. A cavity layout identical to that shown in Figure 2.8 could be used to time-delay and recombine the sweeps. This design could be readily scaled to produce 3 or 4 sweeps per '/2 drive cycle, leading to a 6x or 8x increase in effective sweep rate. Unlike current double-buffered designs, however, no external time delay stages are necessary, leading to a decrease in insertion loss and laser complexity. 2.4 Dispersive FDML Cavities and Sigma Ring Designs The FDML laser designs discussed in previous sections have all operated at centre wavelengths near 1310 nm where chromatic dispersion in SMF-28 singlemode fiber is near zero. Wavelength-dependant variations in fiber birefringence are also low at 1310 nm and are negligible when the cavity length is shorter than -4 km. There are many applications, however, where operation in a dispersive regime is required or when the cavity length must be increased in order to reduce sweep rates. For ophthalmic OCT imaging, the laser must operate at 1060 nm or 800 nm. To use FDML sources with lower-cost commercial OCT platforms, the sweep rate must be reduced to be compatible with moderate digitizer sampling rates. In these situations circular cavity layouts have significant drawbacks. In both of these examples, chromaticity in the birefringent properties of the fiber make it impossible to linearly polarize the entire sweep prior to entering the intra-cavity SOA. A linear polarization is necessary in order to make use of the optimal amplification performance of highly polarized SOA chips. At 1060 nm, additional issues arise due to chromatic dispersion. This section discusses FDML operation in dispersive cavities and introduces sigma-ring geometries to compensate for chromatic birefringence effects. 2.4.1 PolarizationChromaticity FDML lasers exhibit unique and unusual behaviour associated with their polarization properties. Unlike continuous wave or pulsed lasers, where the main problems with polarization management are thermal drift effects, acoustic vibrations and changing stress in the optical components, in FDML lasers an inherent and repeatable change of the polarization state is observed depending on the instantaneous sweep wavelength and position inside the cavity. Immediately after the intra-cavity SOA, which is highly polarized for optimal performance, the entire sweep is linearly polarized parallel to the gain axis of the SOA. Following the long fiber delay and immediately prior to entering the SOA, however, the sweep exhibits wavelength-dependant modulations in polarization state that cannot be removed with simple fiber loop polarization rotators. This chromatic variation in polarization is likely caused by wavelength-dependant behaviour in cavity birefringence linked to the group delay dispersion (GDD) and polarization mode dispersion (PMD) of the fiber. This "polarization chromaticity" is much more severe at wavelengths near 1060 nm, where dispersion in standard single-mode fiber is high, and for low-speed FDML lasers where the cavity is extremely long (> 4 km). Polarization chromaticity makes it impossible to use high-performance polarized SOAs, placing heavy restrictions on performance at 1060 nm. This effect also contributes to deviation of the sweep spectrum away from an ideal Gaussian shape due to polarization dependant loss (PDL) in the laser components, creating sidelobes and broadened main lobes in OCT point spread functions and degrading image quality. Reduction of polarization chromaticity effects can be achieved by applying sigma ring cavity designs, where birefringence is cancelled by use of a Faraday device and double-passing the cavity, as described below. 2.4.2 Dispersive FDML Theory A series of FDML lasers operating at 1060 nm were developed as part of this thesis work in collaboration with Dr. Robert Huber and Dr. Vivek Srinivasan [14, 15]. Operation at 1060 nm or 800 nm is required for ophthalmic imaging due to high water absorption in the eye at 1310 nm. Ophthalmic imaging is the most developed and widespread application of OCT, and is also an application where sample motion is a significant limitation to image quality. Therefore an FDML laser that enables ultrahigh-speed imaging of water-rich samples is highly desirable. Two effects contribute added difficulty to constructing FDML lasers at 1060 nm. The first effect, polarization chromaticity, is discussed above and can be counter-acted with a sigma-ring cavity design. The second effect is group delay dispersion, which is significantly higher at 1060 nm than 1310 nm in conventional singlemode fiber. Dispersive effects can be approximated by generalizing Equation (2.5) to Ardisp = L. DAFW (2.6) where D is the chromatic dispersion coefficient of the fiber in ps/nm/km. The roundtrip timing mismatch due to dispersion ATdisp must be less than the gating time zg over which the filter transmits a given wavelength, as shown in Equation (2.4). This requirement that Adisp < g sets a lower bound on the filter linewidth given by A > c-A2 2D-z A2w D 7 (2.7) nC Equation (2.7) provides several interesting insights for FDML operation within a dispersive cavity. First, it is remarkable that to first approximation the minimum filter bandwidth required for synchronization is independent of the cavity length. A longer cavity causes more dispersion and leads to increased larger temporal desynchronization between the different wavelengths in the sweep. However, the gating time of the filter rg also increases proportionally to cavity length, which largely compensates for this effect. Second, the tuning range A/,F occurs as a quadratic factor in Equation (2.7). This indicates that the filter linewidth must be increased dramatically when a larger tuning range is desired. It is important to emphasize that synchronization for only two optical roundtrips were assumed in this analysis. For true quasi-stationary operation more effective cavity roundtrips and a longer effective cavity photon lifetime are desired, which can be achieved by increasing the filter linewidth beyond the limit described in Equation (2.7). 2.4.3 FDML Lasers at 1060 nm The first 1060 nm FDML laser developed in this project used a standard circular cavity with an external buffering stage, operating at an FFP-TF drive frequency of 118 kHz and an effective unidirectional sweep rate of 236 kHz. The cavity was formed with 1.7 km of Coming HI-1060 singlemode fiber. FDML laser performance was characterized using three FFP-TF filters with different bandwidths, in order to validate the theory developed above. With a dispersion value of D z 40 ps/nm/km for HI-1060 fiber, the refractive index n = 1.46, and a desired tuning range of 80 nm, Equation (2.7) gives a minimum filter linewidth of 0.16 nm for FDML operation. The actual obtained tuning range was 63 nm using a filter linewidth of 0.15 nm, which supported an OCT axial resolution of 15 [tm in air. FFP-TFs with linewidths of -0.08, -0.15, and -0.3 nm were tested and the OCT point spread functions were measured as a function of imaging depth to assess resolution and system sensitivity, as shown in Figure 2.17. From Equation (2.7), these linewidths should support tuning ranges of 56, 77, and 108 nm respectively. Experimentally obtained tuning ranges varied from 60 - 63 nm, indicating that polarization effects (discussed below) were acting to limit tuning range in addition to dispersion effects. The FDML lasers incorporating 0.08 and 0.15 nm filters show comparable sensitivity rolloff at short delays, whereas the 0.3 nm filter has a more rapid rolloff due to a wider instantaneous linewidth. At longer delays the 0.15 nm filter provides slightly enhanced performance compared to the 0.08 nm filter, possibly due to increased losses after one round trip in the case of the 0.08 nm filter. These observations support the dispersive FDML theory and verify that an optimal filter width of -0.16 nm is required for operation near 1060 nm. This laser was used for initial imaging of the human retina at axial line rates of 236 kHz [14]. To increase the tuning range of the laser and support improved OCT axial resolutions, a second 1060 nm laser was constructed with a sigma-ring cavity configuration as shown in Figure 2.18 [15]. The cavity length is 1.65 km and is formed by double-passing a single 825 m section of HI-1060 fiber in the linear portion of the sigma ring, giving an FFP-TF drive frequency of 124 kHz. A fiberoptic Faraday mirror (FM) provides a / 2 polarization rotation prior to the second pass of the 825 m fiber section. This effectively cancels out all birefringence effects in the majority of the cavity except for the short circular portion containing the FFP-TF, polarization controllers (PC), isolators, and SOA. Unidirectional sweeping at 2x the drive frequency is obtained using an external buffering stage identical to that described in Section 2.3.2. Some level of chromatic dispersion compensation was obtained by introducing a Mach-Zehnder interferometer into the cavity with a 15 cm air-path mismatch between the two arms. While the mechanisms of action of the interferometer are not completely understood, it can be thought of as providing two possible paths for photons traveling in the cavity. One path requires an additional 502 ps to travel than the other, and propagating light will travel the path with the lowest total loss in a process similar to mode competition. Light at different wavelengths will therefore self-select a path through multiple roundtrips that optimally matches the filter tuning period, partially correcting for chromatic dispersion effects. Incorporation of the sigma ring cavity and Mach-Zehnder dispersion compensator resulted in a tuning range of 80 nm with a FWHM of 68 nm, compared to a previous tuning range of 63 nm with a FWHM of 38 nm as shown in Figure 2.19. The sigma ring FDML laser supports an OCT axial resolution of 11 tm in air and produced an average output power of 3 mW. This laser was subsequently used for human retinal imaging at a record 248 kHz [15]. Sample 2D and 3D images of a normal human retina are shown in Figure 2.20. Figure 2.20(a,b) show 16,000 axial line images of the fovea (a) and optic nerve head (b) each acquired in 64 ms. Compared to conventional ophthalmic OCT systems operating at 800 nm, the 1060 nm FDML laser enables improved depth penetration into the choroid due to decreased scattering at 1060 nm. Figure 2.20(c,d) show en face reconstructions of a 3D dataset acquired over the fovea (a) and optic nerve head (b). Retinal blood vessels are clearly visible. In Figure 2.20(d), the lamina cribrosa is visible as the large circular feature in the centre of the image. The lamina cribrosa is the structure where the retinal nerves connecting to individual photoreceptors pass into the eye. 2.4.4 Low Speed BroadbandFDML at 1310 nm Although FDML lasers enable extremely high sweep rates and OCT imaging speeds, there are many situations when reduced speeds are desirable. One example can be found in lower-cost commercial imaging systems. These systems, such as the Thorlabs OCS1300SS OCT platform, generally contain lower-speed digitizer cards and data processing software that cannot sustain imaging rates in the hundreds of kHz range. Nevertheless FDML lasers still provide important benefits in terms of low noise, broad tuning, and alignment-free operation that can be beneficial to commercial OCT systems. Desired sweep speeds for systems such as these are generally 20 - 40 kHz, corresponding to cavity lengths of 5 - 10 km. Although chromatic variations in birefringence are much less severe at 1310 nm in SMF-28 fiber, when the cavity length exceeds 4 km noticeable intra-sweep polarization modulations appear in the cavity. These modulations cannot be removed with simple polarization rotator paddles and preclude the use of high-gain, broadband polarized SOAs. Sigma ring cavity designs can be employed in this situation when polarized SOAs are required in a long FDML cavity. To enable high-performance OCT imaging with a commercial Thorlabs data acquisition and processing platform, a sigma ring FDML laser at 1310 nm was constructed as part of this thesis work. The cavity layout is shown in Figure 2.21. The design is similar to the 1060 nm FDML laser shown in Figure 2.18, except the linear fiber section was changed to 4.85 km to give a total cavity length of 9.7 km and an FFP-TF drive frequency of 21 kHz. An external buffer was used to generate unidirectional sweeps at 42 kHz. The fiber was SMF-28 and the intra-cavity interferometer was removed since there is no significant chromatic dispersion at 1310 nm. High-performance polarized SOAs from Covega were use inside the cavity and for extra-cavity booster amplification. Table 2.1 compares the performance of this laser to a buffered circular ring FDML laser with the same cavity length using non-polarized SOA chips. As shown in Table 2.1, the sigma ring geometry provides enhanced performance by enabling the use of polarized SOA chips. Output power is increased by 3.6x, total tuning range is increased by 47%, and axial resolution is improved by 29%. Sensitivity rolloff is slightly worse with the sigma ring design. It is generally true that rolloff performance suffers as the tuning range is increased, likely due to increased instantaneous sweep velocities (nm/s) and therefore reduced gating times at a given sweep rate (Hz). Decreasing the filter gating time rg results in less tolerance to slight mismatches in roundtrip time due to dispersion, lowering the number of effective cavity roundtrips at the edges of the sweep and reducing the instantaneous linewidth. Nevertheless, sigma ring cavity designs have been demonstrated to provide critical performance improvements both for 1060 nm FDML lasers as well as low-speed 1310 nm lasers. Parameter Circular Cavity Sigma Ring SOA Type Non-Polarized / 42 kHz Polarized / 42 kHz Centre Wavelength 1285 nm 1310 nm Average Output Power 13 mW 47 mW Full Width Tuning Range 108 nm 148 nm FWHM Tuning Range 75 nm 101 nm Supported Axial Resolution (Air) 9.7 pm 7.5 pm 6 dB Sensitivity Rolloff 4.5 mm 3.7 mm Table 2.1. Comparison of FDML performance at 42 kHz for a circular cavity with non-polarized SOAs and a sigma ring cavity with polarized SOAs. The sigma ring cavity provides enhanced performance. 2.5 Figures only one RESONATOR or severanl modes are simultanously active in the resonator outcoupler *---forward scan - laser -e-- forward scan - booster .I . 10 .. . d 100 - A . . 1000 . I I ,, 10000 sweep frequency [Hz] Figure 2.1. Conceptual diagram of conventional wavelength-swept lasers. Lasing must build up every time the optical bandpass setting is changed, and only one wavelength is active in the cavity at a given time. Originally published in [7]. Figure 2.2. Output power vs. sweep rate for conventional wavelength-swept lasers both with and without extra-cavity amplification by a booster SOA. First vertical line represents the saturation limit. Second vertical line represents the single roundtrip limit. Output power degrades rapidly as sweep rate is increased. Originally published in [1]. aNmodmrd Figure 2.3. Conceptual diagram of Fourier domain mode-locked (FDML) lasers. The tuning period of the optical bandpass element is synchronized to the cavity roundtrip time. All wavelengths are active in the cavity at all times, removing the fundamental limitation to sweep speed. Originally published in [7]. Figure 2.4. FDML laser schematic showing a circular cavity with 7 km of SMF-28 fiber. Isolators (ISO) ensure that the laser operates in one direction around the cavity. Semiconductor optical amplifiers (SOA) provide gain and a fiber Fabry-Perot tunable filter (FFP-TF) is used as the tunable optical bandpass element. Inset: FDML output is an alternating series of long-to-short and short-to-long wavelength sweeps when a sinusoidal waveform is used to drive the FFP-TF. 290kHz 232kHz 116kHz 3 1 0 -8 4 30 10 20 0 -2 8 4 0 2 4 6 -3-2-10 12345 time [s] Figure 2.5. Transient intensity profiles of an FDML laser operating at 58 - 290 kHz. The profiles remain unchanged as the sweep rate is scaled up, unlike conventional wavelength-swept lasers that degrade with increasing sweep rate. Originally published in [7]. FDML no FDML time [20nsidiv.] time [40ns/div.] 1101 90 0i t 1100 0 0 1 1 20 2 3 4 depth [mm] 5 a 7 1000 20 3000 doph p] Figure 2.6. FDML lasers provide improved stability compared to conventional swept lasers, resulting in narrower instantaneous linewidth and larger ranging depth. Top: time-lapse view of interference fringes acquired with an FDML (left) and conventional swept laser (right). Bottom: OCT point spread functions versus ranging depth for an FDML (left) and conventional swept laser (right). t to OCT 2 6 4 time (~] Figure 2.7. Buffered FDML lasers provide unidirectional sweeps at multiples of the cavity's fundamental frequency. The desired backward sweep is replicated, and the cavity SOA is turned off during the forward sweep. Figure 2.8. Buffered FDML laser schematic for unidirectional sweeping at 370 kHz. The cavity is broken into two sections to create two copies of the backward sweep direction. Sweep copies are recombined in an extra-cavity fiber coupler. 2 depth [mini 3 4 depth [mm] Figure 2.9. Dynamic range versus imaging depth for an FDML laser (left) and a buffered FDML laser (right). The forward sweep in the FDML laser has increased noise compared to the backward sweep, so dynamic range degrades more quickly. In a buffered FDML laser, the forward sweep is eliminated and the backward sweep is replicated, eliminating the dynamic range degradation problem. Originally published in [16]. 2 tS t Figure 2.10. 3D-OCT imaging at 370 kHz using a buffered FDML laser. A: Single cross-sectional image of human skin. 1024 lines acquired in 2.7 ms. B-D: 3D volumetric renderings of human skin. Originally published in [16]. Figure 2.11. Double-buffering concept. The FFPTF is overdriven in amplitude to increase the tuning speed and 4 sweep copies are recombined to quadruple the effective sweep rate relative to the FFP-TF drive frequency. Shaded rectangles indicate times and wavelengths occupied by each sweep copy. OCT O <r) ocO C' Figure 2.12. Double-buffered FDML laser schematic for operation at a sweep rate of 240 kHz with a tuning range of 158 nm at a center wavelength of 1315 nm. Sweep rate is quadrupled by internal and external buffering stages. Originally published in [13]. Figure 2.13. Screen capture of interference fringes and drive waves for a double-buffered FDML laser operating at 240 kHz. Upper trace shows interference signal from an unbalanced Mach-Zehnder interferometer. Middle trace shows the FFP-TF drive waveform. 4 sweeps are generated for each drive cycle. Bottom trace shows the SOA modulation signal, which is ~1/4 the duration of the drive cycle. Point Spread Functions at Various Imaging Depths 0 0 Imaging Depth [mm] Figure 2.14. OCT point spread functions for 240 kHz double-buffered FDML laser measured at increasing imaging depths. Sensitivity rolloff is 6 dB at 2.0 mm. Originally published in [13]. Integrated Laser Spectrum, 240 kHz Double-Buffered FDML Laser 1300 Wavelength [nm] 1400 Figure 2.15. Integrated output spectrum for double-buffered FDML laser, with a total tuning range of 158 nm and a FWHM bandwidth of 117 nm. Originally published in [13]. 2 4 imaging depth [mm] Figure 2.16. Proposed double-buffering concept without external buffering stages. The FFP-TF is overdriven and offset to produce 2 consecutive sweeps over adjacent filter FSR bands during each half drive cycle. Shaded rectangles indicate times and wavelengths occupied by each sweep copy. Figure 2.17. OCT point spread functions for various imaging depths, obtained using an FDML laser at 1060 nm with three different tunable filter linewidths (0.08, 0.15, and 0.3 nm). As filter width increases, a larger tuning range is possible but the instantaneous linewidth is increased and imaging range decreases. Originally published in [14]. 0.8 825 m P I 0.6 to OCT S= 68.32v S 38 i 0.2 A 100 825 m Figure 2.18. 1060 nm FDML laser with sigma ring for polarization chromaticity compensation, intracavity Mach-Zehnder interferometer for dispersion compensation, and external buffering. Sweep rates of 249 kHz with a tuning range of 80 nm are achieved. 1020 1040 ION 1080 1100 1120 wavobr4$ [nm] Figure 2.19. Integrated output spectra for 1060 nm sigma-ring FDML (FWHM = 68 nm) and circular cavity FDML (FWHM = 38 nm). Sigma dispersion intracavity with design ring range tuning improves significantly compensator and supported OCT axial resolution. Figure 2.20. Images of the human retina acquired with a 1060 nm FDML laser. A, high-density crosssection through the fovea. B, high-density cross-section through the optic nerve head. C, en face reconstruction of the fovea. D, en face reconstruction of the optic nerve head. Originally published in [15]. ISO 4.85 km Figure 2.21. 1310 nm FDML laser with sigma ring for polarization chromaticity compensation and external buffering stage. Sweep rates of 42 kHz with a tuning range of 148 nm are achieved using polarized SOA devices. 2.6 References [1] R. Huber, M. Wojtkowski, K. Taira, J. G. Fujimoto, and K. Hsu, "Amplified, frequency swept lasers for frequency domain reflectometry and OCT imaging: design and scaling principles," Optics Express, vol. 13, pp. 3513-3528, May 2 2005. [2] S. H. Yun, G. J. Teamey, J. F. de Boer, N. Iftimia, and B. E. Bouma, "High-speed optical frequency-domain imaging," Optics Express, vol. 11, pp. 2953-2963, Nov 3 2003. [3] W. Y. Oh, S. H. Yun, B. J. Vakoc, G. J. Tearney, and B. E. Bouma, "Ultrahigh-speed optical frequency domain imaging and application to laser ablation monitoring," Applied Physics Letters, vol. 88, pp. -, Mar 6 2006. [4] S. H. Yun, G. J. Tearney, B. J. Vakoc, M. Shishkov, W. Y. Oh, A. E. Desjardins, M. J. Suter, R. C. Chan, J. A. Evans, I. K. Jang, N. S. Nishioka, J. F. de Boer, and B. E. Bouma, "Comprehensive volumetric optical microscopy in vivo," Nature Medicine, vol. 12, pp. 1429-1433, Dec 2006. [5] B. J. Vakoc, M. Shishko, S. H. Yun, W. Y. Oh, M. J. Suter, A. E. Desjardins, J. A. Evans, N. S. Nishioka, G. J. Tearney, and B. E. Bouma, "Comprehensive esophageal microscopy by using optical frequency-domain imaging (with video)," GastrointestinalEndoscopy, vol. 65, pp. 898905, May 2007. [6] R. Huber, K. Taira, M. Wojtkowski, and J. G. Fujimoto, "Fourier Domain Mode Locked Lasers for OCT imaging at up to 290kHz sweep rates," in Optical Coherence Tomography and Coherence Techniques II, Munich, 2005, pp. 245-250. [7] R. Huber, M. Wojtkowski, and J. G. Fujimoto, "Fourier Domain Mode Locking (FDML): A new laser operating regime and applications for optical coherence tomography," Optics Express, vol. 14, pp. 3225-3237, Apr 17 2006. [8] D. C. Adler, R. Huber, and J. G. Fujimoto, "Phase-sensitive optical coherence tomography at up to 370,000 lines per second using buffered Fourier domain mode locked lasers," Opt Lett, vol. 32, pp. 626-628, 2007. [9] R. Huber, D. C. Adler, and J. G. Fujimoto, "Buffered Fourier Domain Mode Locking (FDML): Unidirectional swept laser sources for OCT imaging at 370,000 lines per second," Optics Letters, vol. 31, pp. 2975-2977, October 15, 2006 2006. [10] R. Huber, D. C. Adler, V. J. Srinivasan, and J. G. Fujimoto, "Fourier domain mode locking at 1050 nm for ultrahigh-speed optical coherence tomography of the human retina at 236,000 axial scans per second," Optics Letters, vol. 32, p. In Press, 2007. [11] S. Moon and D. Y. Kim, "Ultra-high-speed optical coherence tomography with a stretched pulse supercontinuum source," Optics Express, vol. 14, pp. 11575-11584, Nov 2006. [12] Y. Park, T. J. Ahn, J. C. Kieffer, and J. Azana, "Optical frequency domain reflectometry based on real-time Fourier transformation," Optics Express, vol. 15, pp. 4597-4616, Apr 2007. [13] D. C. Adler, S. W. Huang, R. Huber, and J. G. Fujimoto, "Photothermal detection of gold nanoparticles using phase-sensitive optical coherence tomography," Optics Express, vol. 16, pp. 4376-4393, Mar 31 2008. [14] R. Huber, D. C. Adler, V. J. Srinivasan, and J. G. Fujimoto, "Fourier domain mode locking at 1050 nm for ultra-high-speed optical coherence tomography of the human retina at 236,000 axial scans per second," Optics Letters, vol. 32, pp. 2049-2051, Jul 2007. [15] V. J. Srinivasan, D. C. Adler, Y. Chen, I. Gorczynska, R. Huber, J. Duker, J. S. Schuman, and J. G. Fujimoto, "Ultrahigh-speed Optical Coherence Tomography for Three-Dimensional and En Face Imaging of the Retina and Optic Nerve Head," 2008, pp. iovs.08-2127. [16] R. Huber, D. C. Adler, and J. G. Fujimoto, "Buffered Fourier domain mode locking: unidirectional swept laser sources for optical coherence tomography imaging at 370,000 lines/s," Optics Letters, vol. 31, pp. 2975-2977, Oct 2006. CHAPTER 3 3.0 Phase Sensitive OCT Using FDML Lasers 3.1 Motivation Contrast in conventional OCT images results from measuring the amplitude of the interference signal formed by a reference field and sample field in a Michelson interferometer. It is also possible to extract the phase of the interference signal, giving an alternative contrast modality that complements the standard amplitude information. Differential (axial line-to-line) phase information can be used to perform Doppler flow OCT [1-6] by detecting variations in interference fringe phase at the same spatial location over short periods of time. It is also possible to map the absolute interference fringe phase relative to a fixed reference surface, enabling the detection of picometer-scale surface displacements or variations in optical path length [7]. These techniques are commonly referred to as OCT phase microscopy or, more generally, phase-sensitive OCT [8-15]. In both cases, small changes in the phase of the interference signal can be correlated to small, sub-resolution changes in the location of backscattering or backreflecting surfaces within the sample. These changes are not detectable using conventional OCT analysis since the phase information is discarded during image formation. The combination of Fourier domain detection and phase sensitive analysis is very attractive, since it enables phase-based contrast at extremely high imaging speeds. The displacement sensitivity of a phasesensitive OCT system is a specification of the smallest variation in optical path length that is detectable by measuring the interference fringe phase relative to a fixed reference surface. The smallest possible displacement sensitivity is desired for detecting low flow rates or small variations in optical path length. Rapid imaging speeds are simultaneously desired to minimize motion artifacts, avoid fringe averaging and 27r phase wrapping artifacts, and enable detection of rapid flows in Doppler OCT applications. OCT phase microscopy systems have been previously reported based on both spectral domain [8-10] and swept source [11] detection. Doppler OCT systems have also been reported using both spectral domain [12-14] and swept source [15] detection. The spectrometer-based systems have provided excellent displacement sensitivities (18 pm at 30 Hz imaging rates [13] or 25 pm at 29 kHz imaging rates [9]) but are limited in speed due to the readout rates of the CCD cameras, limiting the largest flow rates and path changes that can be observed. Prior swept source systems have produced poorer displacement sensitivities (1300 pm at 12 kHz [11] or 475 pm at 16 kHz [15]) due to the low sweep-to-sweep phase stability of conventional swept wavelength laser sources. As part of this thesis work, the phase stability properties of FDML lasers were characterized and compared to conventional swept lasers. Differences in buffered versus non-buffered FDML cavities were observed, along with differences between buffered FDML lasers operating at different sweep rates. High-speed phase sensitive OCT was also demonstrated at axial line rates of a record 117 kHz. 3.2 Phase Stability Measurements of FDML Lasers As discussed above, phase sensitive OCT imaging requires a light source with extremely high phase stability since phase jitter in the laser can easily drown out the phase changes associated with variations in optical path length of the sample. Simultaneously, high-speed imaging is desirable in order to minimize parasitic sample motion and maximize the range of detectable path changes. FDML lasers operate in a quasi-stationary regime where each wavelength in the sweep has many characteristics of a narrowband continuous-wave laser, including high sweep-to-sweep phase stability. The phase of the interference signal generated during OCT imaging is analyzed by Fourier transforming the fringes and then evaluating the phase q (z), as opposed to the amplitude A(z), of the backscattered light as a function of sample depth. 3.2.1 Experimental Setup The experimental setup for performing phase sensitive OCT measurements is shown in Figure 3.1. 95% of the power from the laser source enters a common path Michelson interferometer. In a common path topology, the front surface of a 210 [tm thick glass cover slip provides the reference reflection for the interferometer. A sample, such as a biological cell or photonic device, can be placed on the back surface of the coverslip for interrogation. Common path setups are commonly used in phase sensitive OCT systems to prevent phase jitter between the separate reference arm and sample arm in a standard Michelson interferometer. 5% of the laser output is routed to a second common path interferometer, which is used to record and calibrate the slow phase drift of the FDML lasers. This step is required since the phase noise of FDML lasers consists of a low amplitude (milliradian) white Gaussian component and a larger amplitude (radian) component that drifts slowly. The slow component typically has a period of 1 - 5 ms and is likely caused by thermal drift in the fiber Fabry-Perot tunable filter (FFP-TF) element. Digitization is carried out by a 5 GSample/s 8-bit digital oscilloscope (for displacement sensitivity measurements) or a 200 MSample/s 14-bit digitizer card (for imaging experiments). A personal computer is used for resampling the interference fringes onto a uniform (k) spacing, Fourier transformation, phase extraction, and calibration. Displacement sensitivities Zmin using FDML and conventional swept lasers were measured by recording the phase of the interference signal b(z) originating from the back surface of the 210 gm coverslip (z = z o = 210 pm ) in the sample arm relative to the front surface of the coverslip. The phase 0 (z, ) was measured continuously over -3 ms. The standard deviation of this measurement Uc((zo, was used to calculate displacement sensitivity as [13] t)) Zmin = 4fn (3.1) where A0 is the laser centre wavelength in vacuum and n is the refractive index of the sample material. Differential displacement sensitivity was also measured using each laser source. Differential displacement sensitivity is a measurement of the minimum path length variation that can be detected at the same axial position over two consecutive axial lines. This specification is useful to estimate Doppler OCT performance, since Doppler processing requires differential measurements across time to detect flow. Differential displacement sensitivity Azmin Az where A0(zo,t)= (zo, t + At) - is given by = "- (zo, t)) o(A o47rn (3.2) (3.2) (zo, t) and At is the laser sweep period. It is also possible to calculate the signal-to-noise (SNR) limited displacement sensitivity, which allows comparison of each laser's performance to the theoretical optimum value. SNR-limited displacement sensitivity zSNR is given by [13] ZSNR - n SNR (3.3) where SNR is measured at z = z o = 210pm and is given by SNR = A (zo)2 var (A (z)) (3.4) with A (z) the amplitude of the OCT interference fringe. 3.2.2 Data Processing Since the phase noise of FDML lasers is extremely low, great care must be taken to avoid introducing phase noise artifacts during data processing. Significant phase artifacts can arise from data segmentation, Fourier transformation, and phase extraction. To prevent phase jitter from the PC digitizer board's sample clock and record-to-record rearming time jitter, displacement sensitivity measurements were performed using the 5 GSample/s digital oscilloscope. While rearming jitter is eliminated by storing many interference fringes back-to-back in a single 3 ms long record, individual fringes must be segmented and extracted from the record prior to Fourier transformation. The segmentation technique is a critical data processing step that can easily induce phase errors on the order of the actual FDML laser phase noise. With typical FDML phase noise levels less than 1 mrad at an interference signal frequency of- 10 MHz, for example, data segmentation errors on the order of 100 ps will add considerably to the measured phase noise. 100 ps corresponds to 0.5 samples of a 5 GSample/s signal, indicating a need for custom segmentation software. The data processing flowchart for displacement sensitivity measurements is shown in Figure 3.2. Steps where manual input is required are shown as polygons whereas automated steps are shown as rectangles. Parallelograms indicate steps where data is created or manipulated. Additional figures illustrating the data processing procedure are shown in Figure 3.3 to Figure 3.8. These figures were generated using data acquired with a 21 kHz buffered FDML laser operating at an effective sweep rate of 42 kHz. First, the entire record including the timebase and sample values from the 5 GSample/s oscilloscope is read into Matlab and converted to two double precision vectors. Both the time vector and sample value vectors are roughly 1x10 7 points long and contain 40 - 7400 laser sweeps, depending on the sweep rate. Since the record may start and end partway through a laser sweep, the points containing partial sweeps are removed as shown in Figure 3.3. Next a new timebase is created that will allow perfect segmentation of each sweep. In the original timebase of the oscilloscope, there are generally not an integer number of samples in each laser sweep period. Simple segmentation of the original record would therefore lead to an accumulating error in sweep duration, causing phase walkoff and single-point segmentation jumps that will corrupt the noise measurement. To avoid this, the processing software calculates the smallest number of samples that would be required in the entire record to give an integer number of samples per sweep after segmentation. This requires knowledge of the exact FFP-TF drive frequency fd (which is input by the user), the oscilloscope sampling period At, (which is derived from the timebase vector), and the number of buffering stages in the laser B (if any) to calculate the effective sweep rate. The desired number of samples in the long record Ns, is calculated as N,, = Nces round where Ncycles = floor (NM0o f . At) lfd At, (3.5) is the number of complete FFP-TF drive periods contained in the long record and Nso is the original number of samples in the long record. Once the new timebase is established, the long record is broken up into 4 equal segments in order to avoid memory overflow issues in Matlab. Each segment is then linearly interpolated onto the new timebase and the long record is broken up into segments corresponding to one sweep period. Each new segment is exactly of length N,, / Ncycles . At this point the user must check the segmentation results by examining the first sweep and the last sweep. The sweeps should overlap perfectly to avoid phase walkoff or k-space recalibration errors. If segmentation drift is observed, the user must adjust the FFP-TF drive frequency value and repeat steps 3 - 8 as shown in Figure 3.2. After segmentation, the user must strip the unused "dead points" from the beginning and end of each sweep as shown in Figure 3.4. The process of k-space recalibration can now begin. Here, a recalibration vector is used to resample each interference fringe onto a timebase that is uniformly spaced in optical frequency k. If the laser under test is a buffered FDML laser, each independent path through the buffering stages will generate a slightly different phase evolution. One recalibration vector must therefore be generated for each optical copy that is created (ie, a single-buffered laser with two output couplers in the cavity requires two recalibration vectors, whereas a double-buffered laser requires four recalibration vectors). To generate low-noise fringes for recalibration, all of the fringes are grouped according to buffered output index and then averaged. A 6 th order polynomial fit is subtracted from the averaged fringes in order to reduce the asymmetry about 0 V arising due to unbalanced photodetection. This step is necessary to reduce phase artifacts in the subsequent Hilbert transform, which is used to extract instantaneous phase versus time curves for each calibration fringe. A typical phase versus time curve is shown as the blue trace in Figure 3.5. A 5 th order polynomial fit is then matched to the instantaneous phase curve. Each sample of this polynomial is proportional to the k value of each corresponding sample in the original interference fringes. A linear phase vector is generated through the minimum and maximum values of the polynomial as shown in Figure 3.5. This linear curve represents the desired linear k spacing of each fringe. Spline interpolation is used to resample each segmented interference fringe onto the linear k vector assuming initial k values defined by the polynomial. The resulting fringes have samples that are perfectly linearly spaced in k, giving optimal point spread functions for OCT imaging and preventing phase distortions as the ranging depth is varied. After recalibration each interference fringe is windowed with a Hanning function and Fourier transformed using a fast Fourier transform (FFT) algorithm. The FFT is not zero-padded and does not provide sufficient axial sampling density to accurately track the measured phase over long time periods, since a single point change in the FFT peak location can cause significant phase spikes. The FFT is used only to generate a rough axial amplitude trace of the sample in order to locate the coverslip's rear surface. Following identification of the rear surface, a chirped Z transform (CZT) is used to perform a finelysampled frequency transform centered around the rear surface. The CZT typically contains 213 points and spans a range of +/- 2x the axial position of the coverslip reflection. This gives sufficient axial sampling density to accurately extract phase over long periods of time. The sample phase is measured at the CZT peak corresponding to the coverslip's rear surface in each interference fringe as shown in Figure 3.6. The large peak at sample 2000 is the coverslip surface, while the small peak at sample 4000 is an artifact due to path length ambiguity in the common path geometry. Once the phase has been extracted for each fringe in the dataset, it can be plotted as a function of sweep number as shown in Figure 3.7. If the laser is a buffered FDML configuration, the slight mismatch in optical path length between the two laser output paths results in a small (-2 mrad in this case) constant phase offset between consecutive measurements. This can be corrected by separating the measurement into even and odd data points and then setting the mean of each set to zero. The slight linear component to the phase profile in Figure 3.7 is caused by miniscule segmentation error that slowly builds up over each sweep. This can also be removed by subtracting a linear fit from the data. The final phase measurement is shown in Figure 3.8. This measurement is obtained after subtracting the lowpass-filtered calibration fringe from the separate common-path calibration arm shown in Figure 3.1. The calibration fringe is processed in an identical manner to that described above, except it is lowpass-filtered at the end to remove all phase noise and leave only the long-term phase drift from the laser. The measured standard deviation from the sample fringe is 0.617 mrad over 123 sweeps (3 ms), corresponding to a displacement sensitivity of 42.1 pm in glass with a refractive index of 1.5 at a centre wavelength of 1285 nm. Identical measurements were made, as described below, for a series of lasers including conventional swept sources, non-buffered FDML lasers, and buffered FDML lasers operating at a range of sweep rates. 3.2.3 Conventional Swept Laser To serve as a control in this experiment, a conventional wavelength-swept laser was constructed using a fiber ring, semiconductor optical amplifier (SOA) gain chip, and FFP-TF. The total tuning range was 128 nm at a centre wavelength of 1285 nm with an average output power of 14 mW. The FFP-TF was operated at a drive frequency of 1 kHz, with forward and backward sweeps giving an effective sweep rate of 2 kHz. Output power, tuning range, and phase noise increased significantly at speeds above 2 kHz as described in Chapter 2. The phase noise of the phase sensitive OCT system was measured to be 3.32 mrad using the method described in Section 3.2.1 and Section 3.2.2. This corresponds to a displacement sensitivity of 226 pm, which is comparable to previously-reported displacement sensitivities using conventional swept lasers [11, 15]. The SNR at the back surface of the coverslip was 77 dB, giving an SNR-limited displacement sensitivity of 9 pm. The 25x difference between the actual sensitivity and the SNR limit highlights a major shortcoming of conventional swept lasers. Non-stationary operation leads to greatly increased phase noise, lowering the utility of these sources for phase sensitive OCT. Differential displacement sensitivity was 361 pm, which is roughly a factor of 2 more than the single-line displacement increases sensitivity. This is to be expected, since differential measurements of noisy signals result in in measurement uncertainty. 3.2.4 Non-Buffered FDML Laser A non-buffered FDML laser was constructed by adding a 9.7 km length of SMF-28 optical fiber to the cavity used for the conventional swept laser in Section 3.2.3. The same SOAs, FFP-TF, and fiber output couplers were used in both cases. The FDML tuning range was 112 nm and the average output power was 13 mW at a center wavelength of 1285 nm. The FFP-TF drive frequency was 21 kHz and the effective sweep rate was 42 kHz using both forward and backward sweeps. The measured phase noise r (A (z, t)) was 1.04 mrad, corresponding to a displacement sensitivity of 71 pm. The SNR at the back surface of the coverslip was 68 dB, giving an SNR-limited displacement sensitivity of 26 pm. As expected the differential displacement sensitivity was 115 pm, approximately a factor ofxT higher than the displacement sensitivity. These results illustrate several important features of FDML lasers. First, the quasi-stationary FDML operation leads to significant improvements in phase noise and displacement sensitivity compared to conventional swept lasers. The 42 kHz FDML laser obtained a >3x improvement in displacement sensitivity compared to the conventional swept source, even though the FDML laser was operating at a 41x higher effective sweep rate. Second, the displacement sensitivity of the FDML laser comes within 2.7x of the theoretical SNR limit compared to 25x for the conventional swept laser. This is further evidence that the excess phase noise of FDML lasers is much lower than other swept sources. 3.2.5 Buffered FDML Lasers To investigate differences between buffered and non-buffered FDML lasers, a series of buffered cavities were constructed with fiber lengths of 9.7 km, 3.5 km, and 1.1 km. Effective unidirectional sweep rates were 42 kHz, 117 kHz, and 368 kHz respectively. In all cases the tuning range was > 110 nm and the average output power was > 11 mW at a centre wavelength of 1285 nm. The 42 kHz buffered laser produced a phase noise reading of 0.57 mrad. The displacement sensitivity, SNR-limited displacement sensitivity, and differential displacement sensitivity were 39 pm, 22 pm, and 43 pm respectively. Again, several interesting observations can be made by comparing the buffered and non-buffered 42 kHz FDML laser results. Even though the SNR-limited displacement sensitivities were similar for the buffered and non-buffered lasers (22 pm and 26 pm respectively), the measured displacement sensitivity for the buffered laser was a factor of 1.8x smaller than the non-buffered laser at the same sweep rate. Additionally, the differential displacement sensitivity for the buffered laser was within 10% of the displacement sensitivity. The expected -.2 increase in noise was not observed in the buffered case. This can be explained by the fact that, in buffered lasers, each output coupler produces a time-delayed optical copy of the propagating sweep at evenly-spaced positions in the cavity. The propagating sweep does not undergo amplification or filtering between the output couplers, so the phase correlation between these pair of consecutives sweeps is preserved. This improves the single-measurement displacement sensitivity and also the differential displacement sensitivity, which could significantly enhance Doppler OCT measurements in the future. Similar improvements should be observed in double-buffered FDML lasers. Results for the 117 kHz and 368 kHz buffered FDML lasers, along with the rest of the lasers tested in this section, are shown in Table 3.1. As the sweep rates of the buffered FDML lasers were increased by shortening the cavity length, the displacement sensitivity degraded moderately from 39 pm at 42 kHz to 102 pm at 368 kHz. This compares favorably with previously reported displacement sensitivities of 25 pm at 29 kHz for spectrometer-based phase sensitive OCT systems [9]. However, buffered FDML lasers enable significant increases in data acquisition rate far beyond the speed limits of spectrometer-based systems. The increased speeds possible with buffered FDML lasers could also be used to perform data averaging, further improving displacement sensitivity. In profilometry applications, displacements of a single surface are measured. In this application, buffered FDML lasers would be capable of measuring displacements in a continuous range between the minimum displacement sensitivity (39 - 102 pm) and the laser coherence length (> 4 mm), corresponding to roughly 8 orders of magnitude. The buffered FDML lasers operating at 42 kHz - 368 kHz come within 1.1x - 2.0x of their respective SNR-limited displacement sensitivities. All of the buffered lasers also provided differential displacement sensitivities that are within -10% of the displacement sensitivity, confirming that buffered cavities provide improved sweep-to-sweep phase correlation through extraction of optical sweep copies. Even at a sweep rate of 368 kHz, 184x higher than the conventional swept laser, the buffered FDML still produced >2x lower phase noise and displacement sensitivity than the conventional source. These observations indicate that buffered FDML lasers are ideal sources for high-speed phase-sensitive OCT applications. Laser Type Used Swept Laser (2 kHz) FDML (42 kHz) Buffered FDML (42 kHz) Buffered FDML (117 kHz) Buffered FDML (368 kHz) Displ. Sens. [pm] 226 71 39 52 102 SNR-Limited Displ. Sens. [pm] 9 26 22 38 50 Differential Displ. Sens. [pm] 361 115 43 56 119 Table 3.1. Displacement sensitivity, SNR-limited displacement sensitivity, and differential displacement sensitivity using phase-sensitive OCT with a conventional swept laser, nonbuffered FDML laser, and buffered FDML lasers at several effective sweep rates. Buffered FDML lasers provide superior phase-sensitive performance at the highest possible sweep rates. 3.3 Sub-Nanometer Dynamic Sensing One potential application of high-speed phase-sensitive OCT is to capture small, rapid, transient events to characterize MEMS devices. The buffered FDML laser running at 117 kHz was used to demonstrate dynamic phase-sensitive OCT measurements that may be similar to those required for MEMS characterization. A gold mirror was mounted to a lead-zirconate-titanate (PZT) piezoelectric transducer positioned 1 mm behind a 3 mm glass coverslip in the sample arm. A 3 mm coverslip was used so that the back surface generated a reference reflection while the front surface was outside the beam focus to prevent unwanted interference. The incident beam waist was 15 jim at the 1/e intensity point, giving a Rayleigh length of 138 jim. The PZT was sinusoidally actuated at 5 kHz, while the phase at a single point on the mirror was monitored over a time period of 3 ms. Figure 3.9 shows the measured PZT displacement versus time. The OCT interference fringes were processed in the same manner as described in Section 3.2.2, except that the measured phase at the gold mirror relative to the glass reference surface was plotted as a function of time as the final step. Removal of the residual linear phase (Step 23 in Figure 3.2) would also remove any linear drift in the mirror position relative to the reference surface, which is not of interest for this experiment. The 5 kHz PZT motion is clearly visible in Figure 3.9 as a periodic modulation at +/- 3 nm amplitude. The excellent phase stability of the laser yields a low noise measurement of displacement. The slow modulation at -500 Hz is likely due to motion of the PZT mount relative to the reference surface. This result demonstrates that phase sensitive OCT with buffered FDML lasers can resolve nanometer-scale surface displacements in microsecond time scales. 3.4 Sub-Nanometer 3D Phase Profilometry Another application of high-speed phase-sensitive OCT is to perform 3D measurements of sample optical path length with sub-nanometer precision. This capability, which can be referred to as 3D-OCT phase microscopy, could be useful for material characterization, cellular biology studies, or industrial inspection. 3D-OCT phase microscopy was performed on a 210 jtm glass coverslip by scanning the incident beam over a 1 mm x 1 mm area using a pair of galvanometers. The 3D dataset consists of 230 x 230 axial scans and was acquired in 0.45 s. The phase of the back surface of the coverslip relative to the front surface was measured and displayed as a physical path difference in a false color image. Again, data processing was performed in the same manner as described in Section 3.2.2, except that the measured phase of each axial line was mapped into a spatial location according to the corresponding galvanometer position. The results of this imaging experiment are shown in Figure 3.10. The exceptionally low phase noise of the system results in good displacement sensitivity, enabling visualization of sub-nanometer variations in optical path length. Surface defects consistent with microgrooves are visible on the left hand side of the coverslip. Larger, more homogenous variations in optical path length are visible on the right hand side of the coverslip. These features may be variations in physical sample thickness (+/- 2 nm) or refractive index (+/- 1.4 x 10 ). Small circular features consistent with pockmarks are scattered throughout the sample. This result demonstrates that phase sensitive OCT with buffered FDML lasers can perform 3D surface measurements with nanometer sensitivities at speeds significantly higher than spectrometer-based systems. 3.5 Figures M TRIG CHI CH2 D Conversion ADb Convron slide Calibration Arm PD Amp 2 10 um Amp repsampl FFT phase extraction * phase calibration (CH 1 - Ch 2) Figure 3.1. Common-path Michelson interferometer used for phase-sensitive OCT. Combining the sample and reference paths into a single path reduces phase noise and improves displacement sensitivity. Originally published in [16]. VECTOR TRAES Figure 3.2. Flowchart showing data processing steps for displacement sensitivity measurements and phase-sensitive OCT imaging. is 0 10 20 30 40 70 60 so 90 80 of Figure 3.3. Data processing Step 2. Partial sweeps are removed from the start and end record. ms 2 a of us 90 first the oscilloscope record. Figure shows 5 Wo The [us7- Figure 3.4. Data processing Step 10. Unusable points are removed from each sweep following segmentation. Figure shows one sweep that has been segmented out from the 2 ms record. Figure 3.5. Data processing Steps 14-16. Phase vs. time curves for the original calibration trace (blue) and desired linear phase curve (red). IO FPm PMe Figure 3.6. Data processing Step 21. Sample is extracted at the FFT peak phase corresponding to the glass coverslip's rear surface for each interference fringe. 20 16 10 20 40 so 80 100 120 140 Swep Numbr Figure 3.7. Data processing Steps 22-23. Phase is corrected for path length imbalances in buffered FDML lasers and residual linear phase 0 10 _.2 4 0 20 40 00 o80 100 120 S 14 Sweep Number phase Final 22-24. Steps Figure 3.8. measurement shows structureless noise with a standard deviation of 0.617 mrad, giving a displacement sensitivity of 42 pm. 0.6 1 1.6 Tim [ms] 2 2.6 Figure 3.9. Dynamic measurement of PZT transducer motion using phase-sensitive OCT system and 117 kHz buffered FDML laser. PZT was driven at 5 kHz over +/- 3 nm. Originally published in [16]. 3D OCT Phase Microscopy Innage of 210um Glass Coverslip 0 100 200 Aoo3WO S400 0 S 600 -2 .4 400 3 600 1000 -10 X Distance [uMn] Figure 3.10. 3D-OCT phase microscopy image of a glass coverslip performed using a 117 kHz buffered FDML laser. Sub-nanometer variations in optical path length are visible as small striations, large modulations, and pockmarks. Originally published in [16]. 3.6 References [1] S. Yazdanfar, M. D. Kulkarni, and J. A. Izatt, "High resolution imaging of in vivo cardiac dynamics using color Doppler optical coherence tomography," Optics Express, vol. 1, 1997/12/22 1997. [2] Z. Chen, T. E. Milner, D. Dave, and J. S. Nelson, "Optical Doppler tomographic imaging of fluid flow velocity in highly scattering media," Optics Letters, vol. 22, pp. 64-6, 1997/01/01 1997. [3] Z. Ding, Y. Zhao, H. Ren, J. S. Nelson, and Z. Chen, "Real-time phase-resolved optical coherence tomography and optical Doppler tomography," Optics Express, vol. 10, 2002/03/11 2002. [4] V. X. D. Yang, M. L. Gordon, B. Qi, J. Pekar, S. Lo, E. Seng-Yue, A. Mok, B. C. Wilson, and I. A. Vitkin, "High speed, wide velocity dynamic range Doppler optical coherence tomography (Part I): System design, signal processing, and performance," Optics Express, vol. 11, pp. 794809, Apr 7 2003. [5] V. X. D. Yang, M. L. Gordon, E. Seng-Yue, S. Lo, B. Qi, J. Pekar, A. Mok, B. C. Wilson, and I. A. Vitkin, "High speed, wide velocity dynamic range Doppler optical coherence tomography (Part II): imaging in vivo cardiac dynamics of Xenopus laevis," Optics Express, vol. 11, 2003/07/14 2003. [6] V. X. D. Yang, M. L. Gordon, S. J. Tang, N. E. Marcon, G. Gardiner, B. Qi, S. Bisland, E. SengYue, S. Lo, J. Pekar, B. C. Wilson, and I. A. Vitkin, "High speed, wide velocity dynamic range Doppler optical coherence tomography (Part III): in vivo endoscopic imaging of blood flow in the rat and human gastrointestinal tracts," Optics Express, vol. 11, pp. 2416-2424, Sep 22 2003. [7] M. Sticker, M. Pircher, E. Gotzinger, H. Sattmann, A. F. Fercher, and C. K. Hitzenberger, "En face imaging of single cell layers by differential phase-contrast optical coherence microscopy," Optics Letters, vol. 27, pp. 1126-8, 2002/07/01 2002. [8] M. A. Choma, A. K. Ellerbee, C. H. Yang, T. L. Creazzo, and J. A. Izatt, "Spectral-domain phase microscopy," Optics Letters, vol. 30, pp. 1162-1164, May 15 2005. [9] C. Joo, T. Akkin, B. Cense, B. H. Park, and J. E. de Boer, "Spectral-domain optical coherence phase microscopy for quantitative phase-contrast imaging," Optics Letters, vol. 30, pp. 21312133, Aug 15 2005. [10] H. Li, B. A. Standish, A. Mariampillai, N. R. Munce, Y. X. Mao, S. Chiu, N. E. Alarcon, B. C. Wilson, A. Vitkin, and V. X. D. Yang, "Feasibility of interstitial Doppler optical coherence tomography for in vivo detection of microvascular changes during photodynamic therapy," Lasers in Surgery and Medicine, vol. 38, pp. 754-761, Sep 2006. [11] M. V. Sarunic, S. Weinberg, and J. A. Izatt, "Full-field swept-source phase microscopy," Optics Letters, vol. 31, pp. 1462-1464, May 15 2006. [12] B. R. White, M. C. Pierce, N. Nassif, B. Cense, B. H. Park, G. J. Tearney, B. E. Bouma, T. C. Chen, and J. F. de Boer, "In vivo dynamic human retinal blood flow imaging using ultra-highspeed spectral domain optical Doppler tomography," Optics Express, vol. 11, pp. 3490-3497, DEC 15 2003. [13] M. A. Choma, A. K. Ellerbee, S. Yazdanfar, and J. A. Izatt, "Doppler flow imaging of cytoplasmic streaming using spectral domain phase microscopy," Journal of Biomedical Optics, vol. 11, pp. -, Mar-Apr 2006. [14] H. W. Ren, T. Sun, D. J. MacDonald, M. J. Cobb, and X. D. Li, "Real-time in vivo dblood-flow imaging by movingscatterer-sensitive spectral-domain optical Doppler tomography," Optics Letters, vol. 31, pp. 927-929, Apr 2006. [15] B. J. Vakoc, S. H. Yun, J. F. de Boer, G. J. Tearney, and B. E. Bouma, "Phase-resolved optical frequency domain imaging," Optics Express, vol. 13, pp. 5483-5493, Jul 11 2005. [16] D. C. Adler, R. Huber, and J. G. Fujimoto, "Phase-sensitive optical coherence tomography at up to 370,000 lines per second using buffered Fourier domain mode-locked lasers," Optics Letters, vol. 32, pp. 626-628, Mar 2007. CHAPTER 4 4.0 Photothermal Detection of Gold Nanoshells with OCT 4.1 Motivation Optical coherence tomography (OCT) has been proven to be a powerful tool for assessing tissue architectural morphology. OCT has enabled three-dimensional (3D) imaging of biological samples with micrometer resolutions, and analysis of internal organs has been made possible through the use of minimally invasive miniaturized probes. Although OCT is a very powerful imaging modality it has not been able to leverage recent advances in molecularly-sensitive contrast agents that are revolutionizing other areas of biomedicine by enabling selective interaction between synthetic materials and specific cell types. The synthetic material, such as a pharmaceutical compound, fluorescent molecule, or nanoparticle, is conjugated to a targeting agent such as an antibody or peptide. When the conjugated substance is introduced into a cell culture or organism, it binds only to cells where the targeting agent has high affinity for a specific protein expressed on the cell surface. In this way, a therapeutic or diagnostic agent can be targeted to individual cells exhibiting a certain phenotype or pathology. Molecular targeting technology has led to dramatic advances in drug delivery, fluorescence imaging, and photothermal therapy. Gold nanoparticles, consisting of a silica core and a gold outer shell, are especially attractive for imaging applications due to their customizable absorption and scattering properties, biocompatibility, and ease of conjugation to antibodies and peptides. Gold nanoparticles can be designed to have high absorption at wavelengths where tissue absorption is low, which has also led to their use as photothermal therapy agents. A clear synergy exists between OCT and molecularly-sensitive contrast agents, although it has been challenging to detect low concentrations of these contrast agents using OCT in the past. If these limitations could be overcome, the structure and pathologic state of tissue could be studied in 3D, in vivo, in real time, and with micron-scale spatial resolutions. A technique for performing ultrahigh-speed 3D-OCT with the ability to detect pathologic tissue or cells based on molecular markers would have a profound impact on biomedicine. Development of molecularly specific contrast mechanisms for OCT would fundamentally transform the field by enabling 3D imaging of biological function in addition to architectural and cellular structure. Development of molecular contrast agents for OCT would achieve cross-disciplinary impact by creating a bridge between the OCT, molecular biology and nanotechnology fields. The impact of molecular biology and nanotechnology research would be multiplied by enabling access to new imaging platforms for 3D structural and cellular imaging. The ability to integrate molecular and 3D structural imaging would have powerful applications for small animal studies, accelerating cancer research. As molecular contrast agents become clinically available, this also promises to improve the sensitivity and specificity for early cancer detection in clinical applications. This work will also serve as a launching point for research on other pathologies associated with abnormal protein expressions, such as neurodegenerative and cardiovascular diseases, that have typically not taken advantage of advances in OCT. Methods for more sensitive and specific detection of disease can improve patient outcome and reduce healthcare costs, impacting a major public issue in the United States and the international community. In this section of the thesis work, a novel method for detecting gold nanoparticles using phasesensitive OCT was developed. Photothermal phase modulations were induced in a sample by localized heating with an amplitude-modulated laser diode. These phase modulations were detected with a highspeed phase-sensitive OCT system similar to the one described in Chapter 3. Extremely high signal to noise ratios (SNR) were achieved by detection of the photothermal signal only at the known modulation frequency, similar in effect to lock-in detection, suggesting the possibility of molecularly-specific OCT in the near future. 4.2 Sources of Contrast in OCT OCT produces cross-sectional and 3D images of tissue microstructure by interferometrically measuring the amplitude and echo time delay of backscattered light [1]. OCT imaging can derive contrast from sources that are either endogenous or exogenous to the tissue being imaged. The most commonly utilized source of endogenous contrast is spatial variations in the scattering properties of the tissue, which produces contrast in conventional OCT images. In this case, only the amplitude of the interference signal is analyzed to form the image. Another source of endogenous contrast is velocity or flow within the sample. Typically referred to as Doppler OCT or optical Doppler tomography (ODT), these techniques analyze phase changes in the interference signal over brief time periods to detect vascular blood flow [27]. Endogenous OCT contrast can also be derived from variations in the size of scattering particles in the tissue [8-10] or wavelength-dependant absorption of different tissue components [11, 12] using spectroscopic OCT [13-15]. Finally, non-centrosymmetric endogenous tissue components, such as collagen, can be detected using nonlinear methods such as second harmonic OCT [16-21]. Exogenous contrast agents have not typically been used in OCT, but recently have been studied more closely. As in other biomedical imaging modalities, OCT contrast agents promise to enable enhanced visualization of selected features such as microvasculature, epithelial structures, and diseased or abnormal tissue. Agents such as methylene blue, rhodamine, and indocyanine green can be detected by signatures in their electron relaxation times using pump-probe OCT [22-24]. OCT contrast enhancement has also been demonstrated using scattering microspheres [25], near-infrared (NIR) dyes [26], iron oxide microparticles [27], and, more recently, nanoparticles [28-36]. Gold nanoparticles are particularly attractive contrast agents since they can be targeted to biochemical markers associated with specific types of disease such as cancer [37, 38], which suggests the possibility of highly sensitive and specific OCT detection of early neoplasia. Gold nanoshells consist of an inner silica core surrounded by a thin gold shell. By changing the relative dimensions of the core and shell, the optical resonance frequency of the particles can be tuned from ultraviolet to near infrared wavelengths [28]. This allows customized tailoring of the optical scattering and absorption properties of the particles to suit the needs of the specific application. Gold nanoshells are highly biocompatible, water-soluble, and commercially available. Nanoshells can be designed with high absorption, targeted to cancer cells, and used for photothermal therapy with minimal damage to surrounding tissue [39]. Other types of gold nanoparticles such as nanorods [33] and nanocages [29] exhibit similar properties to nanoshells and can also be used for exogenous OCT contrast enhancement. Methods for detecting exogenous contrast agents using OCT can be divided into two general categories: passive techniques and active techniques. Passive detection techniques rely on time-invariant differences in the optical properties of the agent compared to the tissue to generate contrast. Differences in the absorption of near infrared dyes compared to tissue have been passively detected using spectroscopic OCT [26]. Similar spectroscopic methods have been applied to detect gold nanoparticles, where the absorption of the nanoparticles caused a blue shift of the OCT signal [29]. Optical scattering can also be used to detect gold nanoparticles using conventional amplitude-based OCT [28, 30, 32-34, 36] since the peak scattering or absorption wavelength of the nanoparticles can be selected to overlap with the OCT imaging wavelength. Passive contrast agent detection may be difficult to apply in vivo, however, since the signal is not background-free and variations in the optical properties of heterogeneous tissue can mask the scattering and absorption characteristics of the agents. Active contrast agent detection techniques modulate a property of the agent to enhance visualization against a heterogeneous tissue background. One example of active contrast agent detection is magnetomotive OCT [27, 31, 35]. In this technique, superparamagnetic iron oxide (SPIO) nanoparticles are taken up by cells in the sample tissue and are then exposed to an external magnetic field of 0.06 - 0.5 T that is modulated at 3 - 50 Hz. Modulation of the external magnetic field causes localized motion in regions of the tissue that have taken up the SPIO, and this motion is detected by fluctuations in the amplitude [31] or phase [35] of the OCT interference signal. Magnetomotive OCT achieves a high signalto-noise ratio (SNR) for detecting SPIO nanoparticles since active modulation of the contrast agent results in a detection scheme that is less susceptible to background noise. However, this technique requires the application of fairly strong magnetic field gradients (up to 11 T/m) and is limited in imaging speed due to the relatively slow mechanical response of SPIO-laden tissue. These factors may make magnetomotive OCT challenging to apply for in vivo imaging in humans. 4.3 Imaging Gold Nanoparticles with Photothermal Phase Modulation As part of this thesis work, an active contrast agent detection technique for high-speed OCT imaging based on photothermal modulation was developed and demonstrated. The technique used gold nanoshells designed to have high absorption at 808 nm where tissue absorption is inherently low. A multimode laser diode operating at 808 nm was used to induce small-scale, localized temperature gradients in regions of the sample that contain the contrast agent. These temperature variations altered the optical path length in the sample. Changes in path length were detected using a swept source OCT phase microscopy system [40-43] built using a double-buffered Fourier domain mode locked (FDML) laser operating at 1315 nm and a sweep rate of 240,000 sweeps per second (240 kHz). By modulating the 808 nm laser diode at a known frequency and observing variations in optical path length that occur only at that frequency, the contrast agent can be detected in a way that significantly reduces background noise. Contrast agent SNR's of up to 131 are obtained using modulation frequencies of 500 Hz - 60 kHz. The technique described here can be integrated with 3D-OCT imaging to provide contrast-enhanced images of tissue architectural morphology. In the future, photothermal detection of gold nanoshells using high-speed, phase-sensitive OCT may enable targeted in vivo imaging of disease with high sensitivity and specificity. 4.3.1 ExperimentalSetup A phase-sensitive swept source OCT system, double-buffered FDML laser, and amplitude modulated 808 nm laser diode were used to detect gold nanoshells in this experiment. The setup is shown in Figure 4.1. The swept source OCT phase microscope is similar to the design previously described in Chapter 3 but was modified to collinearly direct the 808 nm laser diode beam and the 1315 nm OCT beam onto the sample. 95% of the FDML output was routed to a common path interferometer, designated as the "sample interferometer." The liquid sample was held in a glass cuvette where the first glass/liquid interface provides the reference reflection for the interferometer. The output of the fiber-pigtailed 808 nm laser diode was combined with the OCT beam using a dichroic mirror. The diode had a maximum continuous wave output power of 300 mW and was pigtailed to a multimode fiber with a 50 [tm core diameter. The 808 nm beam was collimated by a 15 mm focal length lens and the 1315 nm OCT beam was collimated by a 20 mm focal length lens. An XY pair of galvanometer mirrors with a 6 mm clear aperture was used to aim the combined 808 nm / 1315 nm beam on the sample. The beams were focused using a 30 mm focal length achromatic 2 objective lens. The 808 nm beam diameter was -140 Vm at the 1/e intensity point, while the 1300 nm beam diameter was 15 tm at the 1/e2 intensity point. The 808 nm spot size was measured with a CCD camera while the 1300 nm spot size was estimated using a resolution test target. A smaller diameter is generally more desirable for the 808 nm beam in order to increase the energy density and induce larger optical path changes in the sample. However, the larger beam diameter ensured uniform heating in the volume interrogated with the OCT system and also simplified alignment of the 808 nm beam to the OCT beam. The remaining 5% of the FDML laser output was routed to a second common path interferometer, designated as the "calibration interferometer," which used a 210 [m thick glass slide as the sample. The front and back air/glass boundaries generate two fields that interfere to produce a calibration signal for resampling the sample fringes onto a linear k spacing, and for removing slow phase drift caused by the FDML laser [43]. The sample and calibration data were acquired simultaneously using a 2 GS/s, 8 bit digital oscilloscope. Data processing was performed post-acquisition using a personal computer as described in Section 4.3.3. The 808 nm laser diode was modulated by a digital pulse generator that was synchronized to the beginning of each wavelength sweep. The phase noise of the system was measured by placing a 210 [tm glass slide in the sample interferometer and recording the position of the back surface relative to the front surface over 30 ms. The slow component of the phase drift caused by the FDML laser was recorded using the calibration interferometer and subtracted from the sample data as described in Chapter 3. The results of this measurement are shown in the inset of Figure 4.1. The phase noise was measured to be 2.2 mrad, corresponding to a displacement sensitivity of 153 pm. 4.3.2 Sample Preparation Gold nanoshells with a 120 nm diameter core, 16 nm shell thickness, and peak absorption wavelength of 780 nm were obtained commercially. The nanoshells were mixed with deionized water and diluted to a concentration of 1010 mL' (16.6 pM). At this concentration, the nanoshell absorption coefficient at 808 nm was approximately 3.88 cm-' with a FWHM bandwidth of -400 nm. A glass cuvette with a sample path of 200 jim was filled with the solution, and the cuvette was placed in the sample interferometer of the OCT phase microscope. As shown in Figure 4.2, the glass/fluid interface between the cuvette cover and the nanoshell solution was used as the reference reflection for the common path interferometer. The fluid/glass interface between the nanoshell solution and the cuvette body was monitored for small changes in optical path length, corresponding to localized absorption and heating of the nanoshells during exposure to the 808 nm laser beam. For control experiments, pure deionized water was placed in the same cuvette instead of the gold nanoshell solution. 4.3.3 Data Processing Data processing was very similar to the method described for making phase-sensitive measurements in Chapter 3. The interference fringes from the sample interferometer was recorded as a function of time in a single long record using a digital oscilloscope as described above. The interference fringes generated by the calibration interferometer were simultaneously recorded on a separate oscilloscope channel. Data was segmented into individual sweeps and the measured phase as a function of time was extracted using the method described in Chapter 3. The slow long-term phase drift from the FDML laser extracted from the calibration data was low-pass filtered and subtracted from the sample phase. Once the phase response of the liquid phantom sample was obtained, a Fourier transform was applied to measure the induced phase modulations as a function of frequency. 4.4 Thermal Modeling Two separate thermal models were constructed for this portion of the thesis work. The first model was used to estimate the temperature rise in the sample based on the observed phase variations. This model was required since it was impossible to directly measure temperature oscillations in the sample on the order of - 1' C over a sample volume of 3 x 10-3 mm 3 and millisecond time scales. This model was also necessary in order to make maximum temperature rise estimates for biosafety calculations for future in vivo applications. The second model was used estimate the expected thermal response of the sample based on the thermo-optic properties of the sample material and 808 nm laser. This model was used to verify the results of the first model. 4.4.1 Phase to Temperature Conversion The temperature change in the sample is estimated from the measured phase variations by modeling the effects of temperature on optical path length. Although the mechanical and thermal dynamics of the phantom used in these experiments are fairly complex, some understanding of the system behavior can be achieved by modeling two major effects that work to change the optical path length in opposing directions as the sample temperature is increased. First, the refractive index of water decreases with increasing temperature, which tends to decrease the optical path length. Second, the volumetric thermal expansion coefficient of water is positive near room temperature, which tends to increase the optical path length. For the phantom apparatus used in these experiments, the measured optical path length of the sample, z(T), varies with temperature T and can be modeled as: z(T)= L(T)n(T) (4.1) Here, L(T) is the physical path length and n(T) is the refractive index. A change in optical path length Az that occurs due to a change in temperature AT relative to an initial condition T can be written as: Az = z(To +AT)-z(To)= z(To +AT)-zo (4.2) Az = L(To +AT)n(T o +AT)-zo (4.3) Here, zo is the initial optical path length at To . With a volumetric coefficient of expansion p, an initial physical path length Lo, an initial refractive index no , and a variation in refractive index with temperature dn / dT , the change in optical path length can be expressed using: (4.4) L(To + AT)= L o x(1+ jAT) dn dT (4.5) n(T o +AT)= no +-AT Az = Lo(1+ AT) no + -AT dT (4.6) -zo This formulation assumes that the fluid column illuminated by the 808 nm laser is free to expand only in the axial direction. Axial expansion could occur since the cuvette cover was not tightly fixed to the body. Note that the absolute change in optical path length associated with one thermal modulation is < 120 nm for a 500 Hz modulation frequency and < 7 nm for a 60 kHz modulation frequency. These small size scales complicate the dynamic response of the expanding and contracting water column, so the model used here may not precisely reflect the actual behaviour of the system. In this model, dn / dT is assumed to be constant with temperature. The expression for Az can be expanded to give: Cn dn_ Az = Lono + Lo dn AT + LoPATno + LoAT 2 dn dT dT dn dn AT 2 Az = Ldn AT + LonoAT + Lof dT dT The swept source OCT phase microscope measures phase changes Aq (4.7) (4.8) in the OCT interference fringes. A0 is related to Az through the expression: (4.9) Az = -o A 4zC Therefore the estimated sample temperature variation can be calculated by solving the following quadratic expression for AT: +Lon Lodn AT2 + Lo dT dT o AT 4z =0 (4.10) Although it is possible to explicitly solve Equation (10) for AT, in reality P is a function of temperature as well, P=P(T). For water, the variation in P is significant and changes by -50% between 20 and 30C. Therefore Equation (4.10) was solved numerically using a mathematics package. Figure 4.3 and Figure 4.4 show the estimated sample temperature T in the illuminated volume versus 0 the observed OCT signal phase change A0 for a room temperature of 20 C and two different measured -i, To = 20 OC, Lo = 200 lpm, and A0 = 1315 phase ranges. For this model, values of dn/dT = -91 x 10 6 OC 0 -1 nm were used. p values ranged from 207 x 10-6 OC-1 at 20 0 C to 385 x 10 6 oC at 40 C. Since P is several times larger than dn / dT , sample expansion dominates the system and the net optical path length changes is positive. The quadratic term in Equation (4.10) does not contribute substantially to the solution but was taken into account nonetheless. 4.4.2 EstimatedThermal Responses A second model was applied to understand photothermal contrast behavior as the 808 nm power and spot size are scaled. This model enables calculation of a "forward estimate" of thermal response from the thermo-optic properties of the sample and laser. The system was modeled as an absorbing Gaussian cylinder surrounded by an infinite and homogenous medium. This simplified model does not take into account the effects of scattering on the 808 nm beam shape, but provides a reasonable estimate of the relationship between induced temperature change and 808 nm beam parameters. The model also assumes that the spot size is much smaller than the penetration depth due to absorption, which is true for the parameter space considered below. The heat conduction equation in a cylindrical geometry is given by: aAT(t,z,r) at = pa(Z, r) +a pc a2AT(t, z, r) az2 2 + AT(t,z,r) ar 2 + a 2AT(t, z,r) rr 9 (4.11) (4.11) Here, t is time, z is the axial distance from the top of the cylinder, r is radial distance from the center of the cylinder, pa is the absorption coefficient, p is the fluence rate of the laser, p is the density of the medium, c is the specific heat of the medium, and a is the thermal diffusivity of the medium. For small spot sizes relative to the absorption depth, the heat conduction equation is dominated by radial heat transfer [44]. Combined with a cylindrical geometry, this allows the heat conduction equation to be solved in closed form. For one 808 nm modulation period, the temperature variation can therefore be modeled as [44]: AT(t,r=0)=E pc 8a ) In 1 ( , W < 1/Pat < t (4.12) (t-t,,rE= -AT = VE In 1+ W 28 + (a- t t p) W< Pla,t 2 tp (4.13) Here, E is the irradiance, W is the 1/e2 beam radius, and tp is the exposure duration. E and a are calculated from fundamental material properties of the sample, given by: E = 2PLs / trW2 (4.14) a = k /pc (4.15) Here, Pps is the pulse power of the 808 nm laser and k is the thermal conductivity of the surrounding medium. Equations (4.12) and (4.13) were used to model the expected temperature profile for an 808 nm modulation frequency of 500 Hz over one pulse period. The following parameter values were used for the model: p = 1000 kg/m3, c = 4186 J/kg K, t, = 1 ms, Pa = 388 m-', and k = 0.6 W/m K. The value for was obtained from measurements performed by the nanoshell manufacturer. Thermal responses were modeled for 808 nm beam radii of 5 - 40 jtm. The pulse power was chosen to give a constant temperature increase of 1 'C for each beam radius, corresponding to average powers of 2.6 - 18.2 mW at radii of 5 40 [tm. The results of the model are shown in Figure 4.5, with the legend indicating the beam radius and pulse power used to generate each curve. During the second half of each modulation cycle (t = 1 ms to t = 2 ms), the simulation indicates that temperatures decrease to 18% - 62% of their peak values. This suggests that a slow temperature drift may be difficult to avoid. Decreasing the beam radius, however, enables a given thermal increase to be achieved with less incident power, and also results in more rapid cooling. This would result in a larger phase modulation and lower tissue exposure, emphasizing the need to reduce the 808 nm beam diameter for in vivo applications. 4.5 Experimental Results OCT phase microscopy was performed on two samples: one containing pure deionized water and one containing a diluted gold nanoshell solution. In both cases measurements were taken both with and without exposure to the 808 nm laser source. The results of these experiments are shown in Figure 4.6. In Figure 4.6 all plots show the interference fringe phase associated with the second fluid/glass interface, measured over time, at a single spatial location in the cuvette. Figure 4.6(A) shows the measured phase when the sample is pure deionized water and the 808 nm laser is disabled. The phase profile is generally featureless and corresponds to noise. Figure 4.6(B) shows the measured phase when the sample is a 1010 mL' gold nanoshell solution and the 808 nm laser is disabled. Phase noise is increased due to increased scattering in the sample, but no systematic pattern is observed. This nanoshell concentration is consistent with estimates of concentrations that may be attainable in tumor tissue following systemic administration of antibody-labeled nanoshells [32]. Figure 4.6(C) shows the measured phase from the deionized water sample when the 808 nm laser is activated. The 808 nm laser was set to provide 276 mW and was modulated with a 500 Hz square wave with a 50% duty cycle, giving an average power of 138 mW. The 1310 nm FDML laser provided an additional 20 mW of power on the sample. The 808 nm laser modulation pattern is shown at the top of the plot, and the vertical line indicates the time at which the 808 nm laser was switched on at t = 4 ms. No change in the phase is observed compared to Figure 4.6(A), indicating that the absorption of water at 808 nm is not high enough to cause localized heating and induce optical path changes. Figure 4.6(D) shows the measured phase from the nanoshell solution with the 808 nm laser activated. The modulation parameters were identical to those used for the deionized water sample, and the vertical line indicates the time at which the laser was switched on at t = 4 ms. In this case, a strong phase response is observed. The high absorption of the gold nanoshells at 808 nm causes localized heating of the solution, which in turn increases the optical path length of the sample. The phase response of the sample shows the same modulation pattern as the 808 nm laser. Each phase modulation Aq is -1.1 rad peak-topeak, corresponding to a physical path difference AL of -87 nm using AL = ,0 -A0 / 47rn where 0= 1315 nm is the center wavelength of the FDML laser and n = 1.33 is the refractive index of water. Since there is insufficient time for the solution to fully cool using these 808 nm modulation parameters, there is a slow increase in temperature producing a cumulative increase in optical path of -7.5 rad or 590 nm over 30 ms. Although it clear from Figure 4.6 that gold nanoshells can be detected by direct inspection of the measured phase under certain conditions, an enhanced SNR can be achieved by detecting modulations in the phase signal. One approach is to Fourier transform the measured phase and search for a peak at the precisely-known 808 nm modulation frequency. This concept is illustrated in Figure 4.7. The plots in Figure 4.7 show the frequency spectrum of the measured phase for each test condition in Figure 4.6. In each case, the phase data from t = 0 to t = 4 ms was removed. The slow phase increase from gradual heating in Figure 4.6(D) was removed prior to Fourier transformation by subtracting a quadratic fit from the measurement. When the 808 nm laser is disabled (Figure 4.7(A,B)), whether the sample contained deionized water (Figure 4.7(A)) or a nanoshells solution (Figure 4.7(B)), the only characteristic feature of the frequency spectra is I/f noise. The same is true when the 808 nm laser is modulated at 500 Hz but the sample contains deionized water (Figure 4.7(C)). However, when the 808 nm laser is modulated at 500 Hz and the sample contains nanoshells, a strong peak is seen in the frequency spectrum of the measured phase at exactly 500 Hz (Figure 4.7(D)). Smaller harmonic peaks are also visible at 1 kHz frequency increments, consistent with the frequency spectrum of a triangular waveform repeating at 500 Hz. The "signal" in these measurements is defined as the peak Fourier transform amplitude within ± 20 Hz of the nominal 808 nm modulation frequency. This range was selected to allow for a small absolute error in the modulation frequency. The "noise" is defined as the peak FT amplitude within the same ± 20 Hz window measured using the first 4 ms of data. During this time the 808 nm laser was disabled in all cases, allowing for an accurate and consistent estimate of phase noise at the 808 nm modulation frequency. The SNR measurements were repeated five times for each combination of sample type (with and without nanoshells) and 808 nm laser state (disabled or enabled for t > 4 ms). The SNR is defined as the ratio of the signal peak to the noise value. In Figure 4.7 SNR values are shown as the mean of five repeated measurements, plus or minus one standard deviation. As shown in Figure 4.7(A,C), SNR values are insignificant when nanoshells are not present. This indicates the lack of a photothermal modulation signal. SNR values are also insignificant when nanoshells are present but the 808 nm laser is switched off (Figure 4.7(B)). However, when nanoshells are present in the sample and the 808 nm laser is modulated at 500 Hz, the SNR is 131 ± 91. 4.5.1 Signal to Noise Ratio versus Modulation Frequency As the modulation frequency of the 808 nm laser is increased, the frequency of the phase modulation in the nanoshell solution also increases. This has the benefit of shifting the photothermal modulation peak to higher frequencies where 1/f noise is reduced. However, the sample volume illuminated with the 808 nm beam has less time to heat and cool, resulting in a smaller optical path modulation and lower Fourier transform peak amplitude. To investigate this tradeoff three sets of additional experiments were conducted using 808 nm modulation frequencies of 1, 15, and 60 kHz. Each experiment was repeated five times under the same conditions to test for consistency. The results are shown in Figure 4.8. Figure 4.8(AC) shows the measured phase from the nanoshell solution at one transverse position on the cuvette. The vertical line in each plot indicates the time when the 808 nm beam was switched on at t = 4 ms. The insets show enlarged views of 500 ps (Figure 4.8 (B)) and 90 [ts (Figure 4.8(C)) segments of the measured phase with phase modulations clearly visible. For all three cases a gradual path change is visible due to slow heating of the sample, although this effect is lower at 15 and 60 kHz. Figure 4.8(D-F) shows the Fourier transform of the measured phases beginning at 4 ms and with the slow phase component removed by subtracting a quadratic fit. The SNR is measured at each modulation frequency as described in Section 4.5 above. An optimum SNR of 112 ± 45 is achieved at a modulation frequency of 15 kHz. At a modulation frequency of 1 kHz, the SNR suffers from low frequency 1/f noise near the baseband. At a modulation frequency of 60 kHz, the benefits of decreased 1/f noise are outweighed by the decrease in peak modulation amplitude. It is expected that the optimal modulation frequency will vary depending on the optical properties of the sample, nanoshell concentration, and 808 nm laser power level. Although other techniques used to detect gold nanoparticles with OCT have used different phantom systems, which makes direct comparison difficult, reported SNR's have typically ranged from 1.5 - 5 [29, 32-34, 36]. There has been one previous report of SNR values of 79 - 631 using highly backscattering nanoshells in a non-scattering water sample, but the SNR decreased to 5 when a scattering tissue phantom that more closely approximated biological tissue was used [32]. The phantom used in the photothermal experiments described here does not include scattering but accurately reflects the absorption properties of biological tissue [45] and achieves mean SNR's of 2 - 131. The photothermal modulation method is expected to perform well in scattering tissue since the technique is less sensitive to background noise because phase changes are detected at a specific modulation frequency away from baseband. In addition, this technique uses absorption rather than scattering to generate contrast. The high SNR also suggests that lower concentrations of nanoparticles could be detectable using this method than other methods, which is relevant for in vivo applications where the agent is administered systemically and accumulates at lower levels in diseased tissue [34]. 4.5.2 Effect of Measurement Time on Signal to Noise Ratio For in vivo imaging applications, it is important to determine how long one region of the sample must be measured in order to obtain a reasonable contrast agent SNR. Longer measurement times lead to increased SNR but decreased overall frame rates. This tradeoff was evaluated by shortening the time window used in the Fourier transform for an 808 nm modulation frequency of 15 kHz. One representative dataset was used for this test, and the results are shown in Figure 4.9. A linear relationship between the observation time and the SNR is observed in the measured data. As the observation time window is decreased from 28 ms to 2 ms, the SNR decreases linearly from 108 to 7. These results suggest that observation times of only a few ms may be required per transverse position in order to obtain reasonable nanoshell contrast using Fourier transform analysis. Other data analysis techniques may be developed in the future that can obtain similar contrast in shorter time periods. 4.5.3 Comparison to Model Results The phase measurements described in Section 4.5 were compared to the models developed in Section 4.4 in order to estimate the temperature response associated with the measured phase responses in the sample. The estimated temperature response was then compared to the "forward" thermal model based on the thermo-optic properties of the sample and 808 nm laser. Figure 4.10 shows a comparison between the forward model results (blue "expected" curve) and the thermal response estimated from the actual phase measurements shown in Figure 4.6(d) for t = 4 ms to 6 ms. The measured phase was converted to an estimated temperature increase using Equation (4.10). The expected model parameters were adjusted to reflect the actual experimental conditions. A beam radius of 70 tm was used with an average incident power of 138 mW. The expected and estimated modles show good correlation, indicating that the thermal models described here are reasonable. Experimental results showed average phase modulations of ±575, ±369, ±93, and ±32 mrad at 808 nm laser modulation frequencies of 0.5, 1, 15, and 60 kHz, respectively, as shown in Figure 4.6(D) and Figure 4.8. By comparing the measured phase modulations to the model results shown in Figure 4.3, the estimated temperature fluctuations are ±1.47, ±0.98, ±0.26, and ±0.090 C at 808 nm laser modulation frequencies of 0.5, 1, 15, and 60 kHz, respectively. If the 808 nm laser is held at one transverse position for 26 ms, slow phase increases of -4 - 8 rad are observed, corresponding to temperature increases of -7.4 - 12.3 0 C. This slow temperature increase can be minimized in in vivo imaging applications by translating the beam faster, since only a few ms of observation time per transverse location may be required. Additionally, the use of a more tightly focused 808 nm beam would permit more rapid cooling as shown in Figure 4.5. This result also highlights the potential of this technique for conducting photothermal therapy as well as OCT imaging, since even larger temperature gradients could be induced by increasing the 808 nm exposure levels. 4.6 Limitations Although good performance was achieved in this phantom experiment, additional studies are required to demonstrate the utility of the technique in vivo. The phantom used here was selected to test the concept of photothermal detection with the most critical optical property of the sample, absorption, isolated from other effects. Although there are challenges for applying this method in vivo, OCT phase microscopy has previously been applied for studies of living cells and tissue by placing a thin coverslip in contact with the sample [40, 46-48]. Similar approaches may be used for the technique described here, with the top surface of a thin glass window or endoscope sheath in light contact with the tissue providing the reference field. Other groups have demonstrated that optically-induced thermal gradients can result in physical tissue displacements ex vivo that are detectable with phase-sensitive OCT [49]. Although these previous studies induced large temperature gradients and measured correspondingly large phase changes, they indicate that the photothermal modulation contrast mechanism described here should remain valid in tissue using smaller temperature gradients and measuring smaller phase modulations. More fundamentally, recent reports have indicated that it is difficult to accurately measure small phase changes < 0.1 rad in vivo using OCT [50]. This phase uncertainty is due to speckle noise and sample motion, and can be reduced through spatial averaging and other signal processing techniques. With photothermal detection of gold nanoshells, phase variations of + 0.098 rad can be induced at a modulation frequency of 15 kHz and much larger variations are possible at lower modulation frequencies. This indicates that observation of the photothermal modulation signal should be possible in vivo. The necessity to spatially average the phase measurements over the axial or transverse dimensions in order to detect the photothermal modulation could decrease the effective spatial resolution or imaging speed of this technique. It would still be possible, however, to obtain conventional 3D-OCT images at high resolution concurrently with the contrast-enhanced images by applying standard OCT signal processing. Another challenge for in vivo applications is ensuring that the thermal changes induced by the 808 nm laser are small enough to preclude inadvertent tissue damage. The thermal changes have two components: a small-amplitude component associated with the 808 nm modulation over tens of microseconds, and a slow rise in temperature associated with gradual heating over tens of milliseconds. For this discussion, it is assumed that in vivo temperature increases will be similar to those obtained in the phantom experiments. The results shown here indicate that for optimal SNR conditions, the temperature rise associated with the 808 nm modulation is -0.52 OC. This is an extremely small thermal change that is not expected to cause tissue damage. The gradual temperature increase associated with slow sample heating at optimal SNR conditions is -8 'C over 26 ms. For in vivo applications, however, the observation time at one transverse spot could be reduced to 5 ms while maintaining an SNR of -20 which would cause an overall temperature increase of only -3.5 0 C. Since the tissue would only be exposed to this elevated temperature for several milliseconds as the beam is scanned in the transverse dimension, no significant tissue damage is expected. By comparison, photothermal and photodynamic therapies typically require temperature increases of several tens of degrees for many tens of seconds to induce permanent tissue damage. Observation times may be further decreased by developing new analysis algorithms that more efficiently detect the photothermal phase modulation. The use of a single mode laser diode instead of the multimode 808 nm laser diode would allow thermal gradients to be induced over a more localized area. This would reduce the power required to create a given phase modulation and would simultaneously increase cooling rates, further increasing safety margins. One final consideration for in vivo applications is the effect of scattering in biological tissue. Scattering may decrease the effectiveness of the photothermal modulation technique for large tissue depths, since 808 nm light penetration will be reduced. In the phantom experiments performed here, the phase measurement was performed at a fluid/glass interface at a single transverse position. In an in vivo application, the OCT and 808 nm beams would be continuously scanned across the sample in a transverse direction. Due to the spatial distribution of scattering centers in biological tissue, phase measurements would be obtained at each axial position in the sample in a manner similar to Doppler OCT techniques. The photothermal phase modulation may be detected by comparing the measured phase at each axial point to the measured phase in subsequent axial lines, or by comparison to a fixed phase reference such as a glass window or endoscope sheath. The combination of transverse beam scanning and axially distributed phase measurements will add speckle noise to the photothermal signal. Since nanoshell contrast is derived from a phase modulation at a precisely known frequency, however, the effects of random background noise such as speckle and sample motion are expected to be minimized. Increasing the 808 nm modulation frequency may also have important benefits in scattering systems since this will allow more phase modulation periods to be captured over each speckle cell. Higher modulation frequencies could therefore increase OCT frame rates and take maximum advantage of the speed advantage offered by FDML lasers. Although the initial results shown here suggest that the photothermal modulation technique may provide performance benefits in vivo, more study is required to verify this. Future experiments will focus on evaluating SNR performance in solid tissue phantoms that more closely approximate the scattering, thermal conductivity, and mechanical properties of biological tissue. Minimum detectable nanoshell concentrations will be studied in these phantoms and assessed for benefits compared to scattering- or absorption-based contrast. Finally, experiments in in vitro and in vivo biological tissue are needed to validate the photothermal contrast technique. 4.7 Figures System Phase Noise 0.01 0.005 0 -0.005 0.010 10 20 30 Time [ms] Figure 4.1. Swept-source OCT phase microscope with photothermal modulation system. C1, C2, C3, collimating lenses. OBJ, objective lens. DCM, dichroic mirror. X,Y, galvanometer mirrors. PD, photodiode. A, amplifier. TRG, sweep trigger. CH 1, OCT signal input. CH 2, calibration signal input. DAQ, data acquisition. Inset shows measured phase noise of 2.2 mrad. Originally published in [51]. 15 pm II Cuvette Cover Reference Surface II Estimated Temperature vs. Phase Change Nanoshell Solution Cuvette Body \ I Sample Surface I I I 140 pm Figure 4.2. Sample holder and beam geometries for photothermal detection of gold nanoparticles. Beam widths are approximate l/e' points of optical intensity. Originally published in [51]. _0 2 4 6 8 10 Observed Phase Change [rad] Figure 4.3. Thermal modeling results showing estimated sample temperature calculated from observed phase changes from 0 - 10 rad. Originally published in [51]. Modeled Thermal Response, 5-40 Lm Beam Radii Estimated Temperature vs. Phase Change ,O 1.5 1 0.5 Observed Phase Change [rad] '0 2 1 1.5 2 Time [ms] Figure 4.5. Thermal modeling results showing expected temperature increases for 808 nm beam radii of 4 - 40 pm. Originally published in [51]. Figure 4.4. Thermal modeling results showing estimated sample temperature calculated from observed phase changes from 0 - 2 rad. Originally published in [51]. Nanoshells (+) Nanoshells (-) nAu 0.5 a 0.3 -0. 3 51 -0. 10 20 Time [ms] 2 Time [ms] 0.5 C 0.3 0.1 -0.1 -0.3 II Time [ms] 15 Time [ms] Figure 4.6. Measured phase from the back surface of the cuvette for various experimental configurations. 1 laser A: Deionized water with 808 nm laser deactivated. B: 1 x 1010 mL- nanoshell solution with 808 nm laser the shows train pulse Red Hz. 500 at deactivated. C: Deionized water with 808 nm laser modulated Red Hz. 500 modulation signal. D: 1 x 1010 mL-1 nanoshell solution with 808 nm laser modulated at contains sample pulse train shows laser modulation signal. Phase modulations are visible only when nanoshells and when the 808 nm laser is activated. Originally published in [51]. Nanoshells (+) Nanoshells (-) Ln 4) I. 0 0 (U E C Co 0 2000 3000 4000 5000 2000 1000 -0 + 5000 3000 Frequency [Hz] Frequency [Hz] 2 [dl 1500) 1 E SNR = 131 ±91 a E 1000 00 o 0 5001 "' II 3000 2000 Frequency [Hz] 5000 "0 l~n~_A 1000 2000 3000 Frequency [Hz] 1L-' 4000 A 5000 Figure 4.7. Fourier transform of phase vs. time curves shown in Figure 4.6. A: Deionized water with 808 1 nm laser deactivated. B: 1 x 1010 mL- nanoshell solution with 808 nm laser deactivated. C: Deionized 1 water with 808 nm laser modulated at 500 Hz. d, 1 x 1010 mL- nanoshell solution with 808 nm laser modulated at 500 Hz. Strong peak is observed at 500 Hz when nanoshells are present and the 808 nm laser is activated. Originally published in [51]. 15 kHz Modulation 1 kHz Modulation 60 kHz Modulation 1b 8- I 4- 0. 10 0 10 20 Time [ms] Time [ms] 1000. SNR =70t 24 d e 3( Time [ms] SNR=112 a 45 800 600 400 200 -0 2 4 6 8 10 10 12 Frequency [kHz] 14 16 18 Frequency [kHz] 20 35 57 59 61 63 Frequency [kHz] Figure 4.8. Measured phase (A-C) and Fourier transforms (D-F) of measured phase at various 808 nm laser modulation frequencies. Red lines in (A-C) show time when 808 nm laser was activated. Insets in (B,C) show enlarged views of the measured phase. 808 nm laser modulation frequencies were 1 kHz (A,D), 15 kHz (B,E), and 60 kHz (C,F). Originally published in [51]. Estimated and Expected Temperature Increases Signal-to-Noise Ratio vs. Observation Time 0 5 10 15 20 Observation Time [ms] 25 Figure 4.9. Measured gold nanoshell signal-tonoise ratio (SNR) as the sample observation time is increased for an 808 nm laser modulation frequency of 15 kHz. Originally published in [51]. 30 3 Time [ms] Figure 4.10. Comparison of estimated thermal response from phase measurements and expected thermal response from thermo-optic sample properties. Models used a modulation frequency of 500 Hz, beam radius of 70 pm, and average power of 138 mW. 4.8 References [1] D. Huang, E. A. Swanson, C. P. Lin, J. S. Schuman, W. G. Stinson, W. Chang, M. R. Hee, T. Flotte, K. Gregory, C. A. Puliafito, and J. G. Fujimoto, "Optical Coherence Tomography," Science, vol. 254, pp. 1178-1181, Nov 22 1991. [2] S. Yazdanfar, M. D. Kulkarni, and J. A. Izatt, "High resolution imaging of in vivo cardiac dynamics using color Doppler optical coherence tomography," Optics Express, vol. 1, 1997/12/22 1997. [3] Z. Chen, T. E. Milner, D. Dave, and J. S. Nelson, "Optical Doppler tomographic imaging of fluid flow velocity in highly scattering media," Optics Letters, vol. 22, pp. 64-6, 1997/01/01 1997. [4] Z. Ding, Y. Zhao, H. Ren, J. S. Nelson, and Z. Chen, "Real-time phase-resolved optical coherence tomography and optical Doppler tomography," Optics Express, vol. 10, 2002/03/11 2002. [5] V. X. D. Yang, M. L. Gordon, B. Qi, J. Pekar, S. Lo, E. Seng-Yue, A. Mok, B. C. Wilson, and I. A. Vitkin, "High speed, wide velocity dynamic range Doppler optical coherence tomography (Part I): System design, signal processing, and performance," Optics Express, vol. 11, pp. 794809, Apr 7 2003. [6] V. X. D. Yang, M. L. Gordon, E. Seng-Yue, S. Lo, B. Qi, J. Pekar, A. Mok, B. C. Wilson, and I. A. Vitkin, "High speed, wide velocity dynamic range Doppler optical coherence tomography (Part II): imaging in vivo cardiac dynamics of Xenopus laevis," Optics Express, vol. 11, 2003/07/14 2003. [7] V. X. D. Yang, M. L. Gordon, S. J. Tang, N. E. Marcon, G. Gardiner, B. Qi, S. Bisland, E. SengYue, S. Lo, J. Pekar, B. C. Wilson, and I. A. Vitkin, "High speed, wide velocity dynamic range Doppler optical coherence tomography (Part III): in vivo endoscopic imaging of blood flow in the rat and human gastrointestinal tracts," Optics Express, vol. 11, pp. 2416-2424, Sep 22 2003. [8] D. C. Adler, T. H. Ko, P. R. Herz, and J. G. Fujimoto, "Optical coherence tomography contrast enhancement using spectroscopic analysis with spectral autocorrelation," Optics Express, vol. 12, pp. 5487-5501, Nov 1 2004. [9] C. Y. Xu, P. S. Carney, and S. A. Boppart, "Wavelength-dependent scattering in spectroscopic optical coherence tomography," Optics Express, vol. 13, pp. 5450-5462, Jul 11 2005. [10] S. D. Dyer, T. Dennis, L. K. Street, S. M. Etzel, T. A. Germer, and A. Dienstfrey, "Spectroscopic phase-dispersion optical coherence tomography measurements of scattering phantoms," Optics Express, vol. 14, pp. 8138-8153, Sep 2006. [11] D. J. Faber, E. G. Mik, M. C. Aalders, and T. G. van Leeuwen, "Light absorption of (oxy)hemoglobin assessed by spectroscopic optical coherence tomography," Opt Lett, vol. 28, pp. 1436-8, Aug 15 2003. [12] D. J. Faber, E. G. Mik, M. C. G. Aalders, and T. G. van Leeuwen, "Toward assessment of blood oxygen saturation by spectroscopic optical coherence tomography," Optics Letters, vol. 30, pp. 1015-17, 2005/05/01 2005. [13] U. Morgner, W. Drexler, F. X. Kartner, X. D. Li, C. Pitris, E. P. Ippen, and J. G. Fujimoto, "Spectroscopic optical coherence tomography," Optics Letters, vol. 25, pp. 111-113, 2000/01/15 2000. [14] R. Leitgeb, M. Wojtkowski, A. Kowalczyk, C. K. Hitzenberger, M. Sticker, and A. F. Fercher, "Spectral measurement of absorption by spectroscopic frequency-domain optical coherence tomography," Optics Letters, vol. 25, pp. 820-2, 2000/06/01 2000. [15] C. Y. Xu, C. Vinegoni, T. S. Ralston, W. Luo, W. Tan, and S. A. Boppart, "Spectroscopic spectral-domain optical coherence microscopy," Optics Letters, vol. 31, pp. 1079-1081, Apr 2006. [16] C. Vinegoni, J. S. Bredfeldt, D. L. Marks, and S. A. Boppart, "Nonlinear optical contrast enhancement for optical coherence tomography," Optics Express, vol. 12, pp. 331-341, JAN 26 2004. [17] Y. Jiang, I. Tomov, Y. M. Wang, and Z. P. Chen, "Second-harmonic optical coherence tomography," Optics Letters, vol. 29, pp. 1090-1092, MAY 15 2004. [18] B. E. Applegate, C. Yang, A. M. Rollins, and J. A. Izatt, "Polarization-resolved second-harmonicgeneration optical coherence tomography in collagen," Optics Letters, vol. 29, pp. 2252-4, 2004/10/01 2004. [19] S. Yazdanfar, L. H. Laiho, and P. T. C. So, "Interferometric second harmonic generation microscopy," Optics Express, vol. 12, pp. 2739-2745, JUN 14 2004. [20] M. V. Sarunic, B. E. Applegate, and J. A. Izatt, "Spectral domain second-harmonic optical coherence tomography," Optics Letters, vol. 30, pp. 2391-3, 15 Sept. 2005 2005. [21] J. P. Su, I. V. Tomov, Y. Jiang, and Z. P. Chen, "High-resolution frequency-domain secondharmonic optical coherence tomography," Applied Optics, vol. 46, pp. 1770-1775, Apr 2007. [22] K. D. Rao, M. A. Choma, S. Yazdanfar, A. M. Rollins, and J. A. Izatt, "Molecular contrast in optical coherence tomography by use of a pump-probe technique," Optics Letters, vol. 28, pp. 340-342, Mar 1 2003. [23] B. E. Applegate and J. A. Izatt, "Molecular imaging of endogenous and exogenous chromophores using ground state recovery pump-probe optical coherence tomography," Optics Express, vol. 14, pp. 9142-9155, Oct 2006. [24] Z. Yaqoob, E. McDowell, J. G. Wu, X. Heng, J. Fingler, and C. H. Yang, "Molecular contrast optical coherence tomography: a pump-probe scheme using indocyanine green as a contrast agent," Journalof Biomedical Optics, vol. 11, p. 12, Sep-Oct 2006. [25] T. M. Lee, A. L. Oldenburg, S. Sitafalwalla, D. L. Marks, W. Luo, F. J. J. Toublan, K. S. Suslick, and S. A. Boppart, "Engineered microsphere contrast agents for optical coherence tomography," Optics Letters, vol. 28, pp. 1546-1548, Sep 1 2003. [26] C. Y. Xu, J. Ye, D. L. Marks, and S. A. Boppart, "Near-infrared dyes as contrast-enhancing agents for spectroscopic optical coherence tomography," Optics Letters, vol. 29, pp. 1647-1649, JUL 15 2004. [27] A. L. Oldenburg, J. R. Gunther, and S. A. Boppart, "Imaging magnetically labeled cells with magnetomotive optical coherence tomography," Optics Letters, vol. 30, pp. 747-9, 2005/04/01 2005. [28] C. Loo, A. Lin, L. Hirsch, M. H. Lee, J. Barton, N. Halas, J. West, and R. Drezek, "Nanoshellenabled photonics-based imaging and therapy of cancer," Technol Cancer Res Treat, vol. 3, pp. 33-40, Feb 2004. [29] H. Cang, T. Sun, Z. Y. Li, J. Y. Chen, B. J. Wiley, Y. N. Xia, and X. D. Li, "Gold nanocages as contrast agents for spectroscopic optical coherence tomography," Optics Letters, vol. 30, pp. 3048-3050, Nov 15 2005. [30] J. Chen, F. Saeki, B. J. Wiley, H. Cang, M. J. Cobb, Z.-Y. Li, L. Au, H. Zhang, M. B. Kimmey, X. Li, and Y. Xia, "Gold nanocages: bioconjugation and their potential use as optical imaging contrast agents," Nano Letters, vol. 5, pp. 473-7, 2005/03/ 2005. [31] A. L. Oldenburg, F. J. J. Toublan, K. S. Suslick, A. Wei, and S. A. Boppart, "Magnetomotive contrast for in vivo optical coherence tomography," Optics Express, vol. 13, pp. 6597-6614, Aug 22 2005. [32] A. Agrawal, S. Huang, A. W. H. Lin, M. H. Lee, J. K. Barton, R. A. Drezek, and T. J. Pfefer, "Quantitative evaluation of optical coherence tomography signal enhancement with gold nanoshells," Journalof Biomedical Optics, vol. 11, p. 8, Jul-Aug 2006. [33] A. L. Oldenburg, M. N. Hansen, D. A. Zweifel, A. Wei, and S. A. Boppart, "Plasmon-resonant gold nanorods as low backscattering albedo contrast agents for optical coherence tomography," Optics Express, vol. 14, pp. 6724-6738, Jul 2006. [34] A. M. Gobin, M. H. Lee, N. J. Halas, W. D. James, R. A. Drezek, and J. L. West, "Near-infrared resonant nanoshells for combined optical imaging and photothermal cancer therapy," Nano Letters, vol. 7, pp. 1929-1934, Jul 2007. [35] J. Oh, M. D. Feldman, J. Kim, H. W. Kang, P. Sanghi, and T. E. Milner, "Magneto-motive detection of tissue-based macrophages by differential phase optical coherence tomography," Lasers in Surgery and Medicine, vol. 39, pp. 266-272, Mar 2007. [36] T. S. Troutman, J. K. Barton, and M. Romanowski, "Optical coherence tomography with plasmon resonant nanorods of gold," Optics Letters, vol. 32, pp. 1438-1440, Jun 2007. [37] C. M. Pitsillides, E. K. Joe, X. B. Wei, R. R. Anderson, and C. P. Lin, "Selective cell targeting with light-absorbing microparticles and nanoparticles," Biophysical Journal, vol. 84, pp. 40234032, Jun 2003. [38] K. Sokolov, M. Follen, J. Aaron, I. Pavlova, A. Malpica, R. Lotan, and R. Richards-Kortum, "Real-time vital optical imaging of precancer using anti-epidermal growth factor receptor antibodies conjugated to gold nanoparticles," Cancer research, vol. 63, pp. 1999-2004, May 1 2003. [39] C. Loo, A. Lowery, N. Halas, J. West, and R. Drezek, "Immunotargeted nanoshells for integrated cancer imaging and therapy," Nano Letters, vol. 5, pp. 709-711, Apr 2005. [40] M. A. Choma, A. K. Ellerbee, C. H. Yang, T. L. Creazzo, and J. A. Izatt, "Spectral-domain phase microscopy," Optics Letters, vol. 30, pp. 1162-1164, May 15 2005. [41] C. Joo, T. Akkin, B. Cense, B. H. Park, and J. E. de Boer, "Spectral-domain optical coherence phase microscopy for quantitative phase-contrast imaging," Optics Letters, vol. 30, pp. 21312133, Aug 15 2005. [42] B. J. Vakoc, S. H. Yun, J. F. de Boer, G. J. Tearney, and B. E. Bouma, "Phase-resolved optical frequency domain imaging," Optics Express, vol. 13, pp. 5483-5493, Jul 11 2005. [43] D. C. Adler, R. Huber, and J. G. Fujimoto, "Phase-sensitive optical coherence tomography at up to 370,000 lines per second using buffered Fourier domain mode-locked lasers," Optics Letters, vol. 32, pp. 626-628, Mar 2007. [44] M. J. C. van Gemert, G. W. Lucassen, and A. J. Welch, "Time constants in thermal laser medicine .2. Distributions of time constants and thermal relaxation of tissue," Physics in Medicine and Biology, vol. 41, pp. 1381-1399, Aug 1996. [45] T. L. Troy and S. N. Thennadil, "Optical properties of human skin in the near infrared wavelength range of 1000 to 2200 nm," JournalofBiomedical Optics, vol. 6, pp. 167-176, Apr 2001. [46] M. A. Choma, A. K. Ellerbee, S. Yazdanfar, and J. A. Izatt, "Doppler flow imaging of cytoplasmic streaming using spectral domain phase microscopy," Journal of Biomedical Optics, vol. 11, pp. -, Mar-Apr 2006. [47] A. K. Ellerbee, T. L. Creazzo, and J. A. Izatt, "Investigating nanoscale cellular dynamics with cross-sectional spectral domain phase microscopy," Optics Express, vol. 15, pp. 8115-8124, Jun 2007. [48] E. J. McDowell, A. K. Ellerbee, M. A. Choma, B. E. Applegate, and J. A. Izatt, "Spectral domain phase microscopy for local measurements of cytoskeletal rheology in single cells," Journal of Biomedical Optics, vol. 12, pp. -, Jul-Aug 2007. [49] B. J. Vakoc, G. J. Tearney, and B. E. Bouma, "Real-time microscopic visualization of tissue response to laser thermal therapy," JournalofBiomedical Optics, vol. 12, p. 3, Mar-Apr 2007. [50] S. Makita, Y. Hong, M. Yamanari, T. Yatagai, and Y. Yasuno, "Optical coherence angiography," Optics Express, vol. 14, pp. 7821-7840, Aug 2006. [51] D. C. Adler, S. W. Huang, R. Huber, and J. G. Fujimoto, "Photothermal detection of gold nanoparticles using phase-sensitive optical coherence tomography," Optics Express, vol. 16, pp. 4376-4393, Mar 31 2008. CHAPTER 5 5.0 3D-OCT Platform for Clinical Gastroenterology 5.1 Motivation Clinical environments require a degree of reliability and robustness in imaging technology that presents challenges to translating research from the laboratory bench to the patient's bedside. For OCT imaging systems, real-time data acquisition and display are needed to provide continuous feedback to the operator. Reliability, ease of use, and rapid data acquisition are also required to maintain short procedure times and efficient patient flow. For these reasons, typical benchtop or commercial 3D-OCT systems are not suitable for clinical use and must be heavily modified prior to conducting human studies in a clinical environment. Customized OCT systems have previously been applied to numerous biomedical fields including ophthalmology [1], cardiology [2], gastroenterology [3], urology [4], and gynecology [5]. One particularly promising application for ultrahigh speed 3D-OCT imaging is gastroenterology. Previously, two-dimensional OCT (2D-OCT) imaging of the gastrointestinal (GI) tract has been demonstrated with high resolution but limited imaging speeds of 0.25 - 20 kHz [6-8]. 3D-OCT would provide a significant advantage since it allows a greater region of tissue to be analyzed and reduces the risk of failing to detect an abnormal lesion. Additionally, 3D-OCT enables a wide range of image analysis and visualization techniques to be applied and results in a more complete characterization of tissue microstructure. 3DOCT imaging of the esophagus has been previously demonstrated using a Fourier domain OCT system based on a conventional wavelength-swept laser source [9]. The imaging rate was limited to 10 kHz, however, and the resolvable transverse feature size was restricted to 64 ipm due to coarse spatial sampling. Since the earliest GI lesions typically appear as structures that are -100 [tm in diameter [10], this performance is not suitable for detecting early cancers. Furthermore, the slow imaging rate resulted in a long acquisition time (5.8 minutes) and the presence of significant motion artifacts. In this section of the thesis work an FDML laser was interfaced to a prototype high-speed data acquisition and data processing system built by Joseph Schmitt at LightLab Imaging Inc. This system enabled 3D-OCT clinical studies in the GI tract at the unprecedented imaging speeds supported by FDML technology. The laser and data processing system were modified to provide an optimal combination of imaging speed, ranging depth, axial resolution, and sensitivity. This technology development effort produced a 3D-OCT endomicroscopy platform that was used to support clinical pilot studies of a range of human pathologies. The system will continue to serve this purpose for the foreseeable future. 5.2 System Description The system constructed here was a swept-source OCT configuration using an FDML laser as the light source. Data acquisition, real-time processing, and display was carried out by a computer system designed by LightLab Imaging Inc. Post-processing and image visualization was performed using a combination of purpose-built Matlab software and a commercial 3D rendering package (ResolveRT, Mercury Computer Systems). The imaging probes used to conduct clinical studies are a critical aspect of the system and are described fully in Chapter 6. Preliminary 3D-OCT data used to guide development of the system, signal processing, and visualization methods was acquired using a modified commercial cardiovascular probe, also described in Chapter 6. To be practical in a clinical setting a 3D-OCT endomicroscopy system must provide continuous realtime data capture and display with dense spatial sampling. Extremely large datasets must be digitized, transferred to computer memory, processed, and displayed in milliseconds. The OCT imaging engine developed here operated at sustained speeds compatible with FDML lasers by using a novel optical frequency clock (OFC) integrated with a 200 Msample/s 12 bit analog-to-digital converter (ADC). The OFC circuitry was designed and constructed by LightLab Imaging Inc. OFC optimization for use with an FDML laser was carried out as part of this thesis work as described in Section 5.4. The formation of OCT images from frequency domain interference fringes requires the interference signal to be evenly spaced in optical frequency v prior to Fourier transformation[11, 12]. Since the frequency sweeps of both conventional and FDML lasers are not linear in time, it is necessary to correct for this effect. One common approach is to acquire a reference interference fringe simultaneously with the OCT signal. The v evolution of the reference signal can be analyzed and used to later correct the v spacing of the OCT signal[l 1, 13]. This has the disadvantages of doubling the required DAQ bandwidth since two signals must be captured, requiring signal post-processing to correct the OCT data and limiting real-time operation. The OCT imaging engine developed here used an optical triggering technique to automatically correct for nonlinear v spacing during data acquisition without storing a separate reference signal. Figure 5.1 shows a schematic of the system. 5% of the laser output was routed to an asymmetric Mach-Zehnder interferometer (MZI) that produces interference fringes with zero crossings evenly spaced in v. This concept is shown in the left inset of Figure 5.1. The MZI fringes are detected by a dual balanced photoreceiver and the zero crossings are identified by an analog voltage comparator in the clock generator, creating a digital pulse train that functions as an OFC. The OFC is synchronized to the start of each laser sweep by a trigger signal from the FDML laser. A dual-balanced Michelson interferometer consisting of a pair of optical circulators and a 50/50 fiberoptic splitter is used to generate the OCT interference fringes. Dual-balanced detection cancels excess noise, reducing the dynamic range requirements of the DAQ system and improving sensitivity. The same frequency sweep from the laser generated both the MZI and OCT fringes, so the spacing of the OFC pulses corresponded to evenly spaced v intervals in the OCT signal. The OFC triggers the 12 bit, 200 MSample/s, circularly buffered ADC that samples the OCT fringes. Since the resulting signal is evenly spaced in v, only one signal needed to be digitized and stored and no post-processing was required for v correction. The sample arm of the Michelson interferometer included a Patient Interface Unit (PIU), supplied by LightLab Imaging Inc., that produced high-speed rotational motion at up to 80 Hz and linear pullback motion at 0.5 - 2.0 mm/s. The PIU attached to the proximal end of the imaging probe and produced a cylindrical spiral scan at the distal tip of the probe. After digitization the OCT signal was continuously streamed to the computer RAM over a PCI-X link at an average of 46 MB/s and a peak of 150 MB/s. Hamming windowing and a fast Fourier transform were then performed in software to synthesize the image. Each frame was interpolated into polar coordinates and displayed as a radial image in real-time at > 20 frames/s. During 3D acquisition, sustained acquisition rates of 100 kHz were achieved while maintaining real-time display. 5.3 Laser Design and Optimization The FDML laser source was a key component of the 3D-OCT endomicroscopy system. Laser performance directly affects imaging performance and therefore careful selection of each FDML design parameter was necessary to ensure optimal clinical utility. Figure 5.2 shows a schematic diagram of the non-buffered FDML laser used in the 3D-OCT endomicroscopy system. The cavity length was 3.4 km, giving an FFP-TF drive frequency of 59 kHz and a bi-directional sweep rate of 118 kHz. Only the backward sweep was used to acquire data, however, giving an effective sweep rate of 59 kHz. It is possible to operate the system at effective sweep rates of up to 100 kHz but, as described in Section 5.3.1, a lower sweep rate resulted in optimal imaging performance for the human GI tract. It is also possible to use a buffered FDML laser to obtain unidirectional sweeps at a high duty cycle, but due to the data acquisition limitations described in Section 5.3.2 a non-buffered laser was the more optimal design. The fiber Fabry-Perot tuneable filter (FFP-TF) used in the FDML laser was acquired from LambdaQuest. The FFP-TF had a finesse of -600 and a free spectral range (FSR) of -190 nm, giving a linewidth of -0.3 nm. The semiconductor optical amplifier (SOA) chips were high-gain, broadband, polarization-sensitive devices providing a gain bandwidth of > 100 nm (full width at half maximum, FWHM) and amplified spontaneous emission (ASE) power of > 5 mW. The output coupling ratio was chosen to provide maximum tuning range while maintaining a post-booster output power that was sufficiently high to reach the tissue exposure limit of 20 mW after transmission through the OCT interferometers and imaging probe. With a 50% output coupler, the total tuning range was 180 nm and the average output power after booster amplification was - 50 mW. Power output directly from the cavity was -5 mW. Figure 5.3 shows a series of point spread functions (PSF) measured by connecting the FDML laser to an unbalanced MZI and varying the path imbalance. Data was collected using a digital oscilloscope and only the low-noise backward sweep was used for analysis. Sweep recalibration and Fourier transformation was performed offline using Matlab software with the method described in Chapter 3. A reference fringe at an MZI delay of -250 Rm was used to recalibrate all of the other fringes in the dataset. It is generally not advisable to use a long MZI delay (> 1 mm) to generate reference fringes for recalibration due to decreased coherence length and fringe contrast at larger path imbalances. Use of extremely short delays (< 100 ptm) is also not advisable due to an insufficient number of fringe cycles to calculate an accurate phase evolution curve. This measurement gives the axial resolution and sensitivity rolloff supported by the laser without taking into account the effects of the OFC and real-time data processing algorithms used in the complete 3D-OCT endomicroscopy system. Sensitivity rolls off by 6 dB at an imaging depth of 1.6 mm in air and by 10 dB at 2.8 mm in air. Figure 5.4 shows an enlarged view of a single PSF at an imaging depth of 65 gim. The FWHM of the PSF is 6.4 tm on a linear scale, corresponding to an axial resolution of 4.6 gm in tissue. It should be noted that the sensitivity rolloff and axial resolution specifications for the entire system are slightly degraded compared to the values supported by the laser itself, as described in Section 5.5, due to the effects of the OFC and real-time data processing methods. 5.3.1 Sweep Rate The sweep rate of the FDML laser directly determines the imaging speed of the 3D-OCT endomicroscopy system. Although higher imaging speeds are desirable to maximize the field of view and minimize motion artifacts, imaging speed also trades off against ranging depth and sensitivity. As the sweep rate is increased, the frequency associated with the interference fringe generated at a fixed imaging depth increases in direct proportion. Since the 3D-OCT endomicroscopy system has a fixed ADC with a maximum sampling rate of 200 MSamples/s, the maximum interference fringe frequency that can be sampled without aliasing is 100 MHz. From Equation (1.1) in Chapter 1, the interference fringe signal idet (t) generated at the photodetector for a fixed point reflector at zo is: det (t) cos(2k(t)z +0) (5.1) If the FDML sweep is linear in k between values ko and k, ,bidirectional, and has a backward sweep duration of Tback , then the variation k (t) for the backward sweep only is: (k, -k)(5.2) k(t) = ko + T (5.2) t back The photocurrent can then be rewritten as: idet (t) S 2z (k,-k Tback cos 2koz o + This gives the interference fringe frequency fdet = fdet = de ) (5.3) (5.3) + / 2r as: (5.4) z(k, - ko ) CTback The FDML laser used in the 3D-OCT endomicroscopy system tuned between 1225 nm and 1405 nm. 6 This gives values of 4.47x10 6 m -' and 5.13x10 m ' for ko and k, respectively. The maximum allowable fringe frequency of 100 MHz therefore occurs at z o =476[m/s]-Tback. If the backward sweep is assumed to occupy 50% of the total FFP-TF drive period, a drive frequency of 59 kHz gives a fringe frequency of 100 MHz at a range of 4 mm in air. Since most applications require a ranging depth of at least 1.5 mm in tissue in addition to the 1.25 mm radius of the imaging probe (total range of 3.8 mm in air), an effective sweep frequency of 59 kHz is a good operating point. The number of k samples acquired during each sweep can be estimated as N = 2 Tback fclk where flk = fet (z 0 = Zmax ) is the OFC fringe frequency at an MZI delay of Zmax equal to the maximum ranging depth. Note that this is a rough approximation and assumes a linear sweep with no wavelength dependence in coherence length. As discussed in Section 5.4, there is increased loss of fringe contrast at the edges of the sweep which significantly reduces the number of acquired points at large MZI delays. Sensitivity is also affected by sweep rate, as shown in Equation (1.6) in Chapter 1. Assuming a detector efficiency of 50%, incident power on the sample of 20 mW, and centre wavelength of 1315 nm, the theoretical sensitivity for a 59 kHz filter drive frequency and a 50% backward-sweep duty cycle is 118 dB in the shot noise limit. In reality, excess loss in the sample arm and Michelson interferometer can be expected to reduce the sensitivity by -5 dB. Imaging in highly-scattering biological tissue such as GI mucosa requires higher sensitivities than are typically necessary in ophthalmic OCT applications. Sensitivities of -100 dB or better are necessary to obtain high-quality images, so 59 kHz was also found to be a good operating point from a sensitivity perspective. 5.3.2 Buffered versus Non-Buffered Cavity Selection As described in Chapter 2, buffered FDML lasers are capable of generating unidirectional wavelength sweeps with overall duty cycles near 100%. Buffered FDML lasers are excellent choices for data acquisition systems where little or no "dead time" is required between consecutive wavelength sweeps to perform data acquisition or signal processing steps. Many non-real-time imaging systems would be examples of this situation, since they can acquire one long record and perform sweep segmentation offline following conclusion of the data acquisition. The 3D-OCT endomicroscopy system uses real-time data processing and image display in order to provide maximum utility to the clinical user. The system requires -4.5 [ts between the end of one sweep and the beginning of the next sweep to perform data offload between the ADC and the computer, conduct signal processing, and display the axial line. Because of this requirement for significant dead time, a buffered FDML design is sup-optimal since the high duty cycle could not be fully utilized. A nonbuffered FDML design was more efficient since the noisy forward sweep could be removed via SOA modulation and the resulting time window could be used to perform the ADC and processing tasks. The choice of a non-buffered FDML laser also increased the output power by -2x compared to a buffered laser due to use of a single output coupler and removal of the external 50/50 splitter prior to the booster SOA. The 59 kHz non-buffered FDML laser was initially operated at an overall duty cycle of 90% (ie, laser output was produced during 90% of the FFP-TF drive period) split evenly between the forward and backward sweeps. This provided 7.5 jis forward and backward sweeps. After removal of the forward sweep, the dead time between consecutive backward sweeps was 9.2 [ts. Application of a non-sinusoidal drive waveform, as described in Section 5.3.3, was used to reduce the dead time to 6.9 [ts by temporally compressing the forward sweep. This provided enough dead time for ADC and signal processing tasks and allowed for a reasonable safety margin. 5.3.3 Sweep Linearization The FFP-TF of an FDML laser can be driven with a non-sinusoidal waveform in order to temporally shape the output sweep. This method can be used to generate a sweep that is linear in optical frequency v as described by Eigenwillig et al [14]. Additionally, the drive wave can be skewed to generate a backward sweep that is longer in duration than the forward sweep. This has the advantage of increasing exposure time during the low-noise backward sweep, increasing sensitivity. Sweep skewing also enables the intersweep dead time to be optimized for use with the 3D-OCT endomicroscopy system. Finally, increasing the duration of the backward sweep decreases the interference signal frequency associated with a given ranging depth. This effectively increases the maximum imaging depth that can be interrogated with a given ADC sampling rate. The FDML laser used in this section of the thesis work was linearized and skewed to favour the backward sweep. First, an ideal FFP-TF displacement waveform was generated by choosing a skew ratio and linearity region. A 2:1 skew ratio was chosen (ie, backward sweep occupying twice as much time as the forward sweep) to reduce the inter-sweep dead time to approximately 7 gs for an FFP-TF drive frequency of 59 kHz. For applications in the 3D-OCT endomicroscopy system, sweep linearity was only required during the backward sweep. The forward sweep was not used and therefore the exact shape of the forward section of the FFP-TF drive wave was not important. Next the frequency response of the FFP-TF was measured by applying a low-amplitude sine wave to the filter, scanning the frequency, and recording the amplitude and phase of the FFP-TF displacement. An SOA generating broadband ASE was connected to the FFP-TF input. A portion of the FFP-TF output was directed to an optical spectrum analyzer (OSA). The width of the transmitted spectrum was used to measure the amplitude of the FFP-TF response at each drive frequency. The phase of the response was measured by directing the rest of the FFP-TF output to an unbalanced MZI and observing shifts in the start time of observed interference relative to the start of the driving sine wave. Figure 5.5 shows the measured amplitude response of the FFP-TF used in the FDML laser. Clear harmonic peaks are seen near 56 kHz and 163 kHz. To avoid applying excess electrical power to the FFP-TF, the ideal drive wave was approximated by a summation of three harmonic sine waves at 59 kHz, 118 kHz, and 177 kHz. These frequencies are near the resonant peaks of the filter but are not directly on resonance. Driving directly on resonance can lead to unstable behaviour and filter damage. The sinusoidal fit was carried out using Matlab curve fitting utilities. The forward portion of the drive wave was weighted at 0% to allow maximum accuracy for the backward portion. The resulting ideal drive wave is shown in Figure 5.6. This wave represents the desired response of the FFP-TF to give a linear backward sweep with a 2:1 skew ratio. This desired response is modified according to the measured amplitude and phase responses at the three harmonic drive frequencies as shown in Figure 5.5. This correction results in the actual drive waveform shown in Figure 5.7. This drive wave is applied to the filter with minor manual adjustments used to minimize the PSF width. The results of the drive linearization are shown in Figure 5.8 and Figure 5.9. Figure 5.8 shows the amplitude and phase output of an unbalanced MZI using an FDML laser with a linearized, 2:1 skewed waveform driving the FFP-TF. This data was acquired using a digital oscilloscope. The overall duty cycle is 88% with the remaining 12% used as a buffer to prevent sweep overlap during DC bias setpoint drift. The backward sweep, which is used to generate images in the 3D-OCT endomicroscopy system, is > 2x longer than the forward sweep. The phase evolution of the backward sweep is much more linear than the forward sweep, although the backward sweep deviates from a linear phase evolution near the end of the drive period. Figure 5.9 shows Fourier transformed PSF's generated at the same imaging depth using different FFP-TF drive waves. The blue curve is an "ideal" PSF generated by recalibrating a sinusoidal wavelength sweep with post-processing software to give a maximally linear v evolution. The red curve is a non-recalibrated PSF generated with a linear drive wave but no skewing. The PSF width is similar to the ideal case with a low-frequency tail extending down to 15 MHz. The black curve is a non-recalibrated PSF generated with a sinusoidal drive wave. The peak RF frequency is 22% higher than the PSF from the linearized waveform and the bandwidth is -4x larger, indicating inefficient use of ADC time and unnecessarily high peak RF frequencies. The magenta curve is a non-recalibrated PSF generated with a linearized and skewed drive wave. The RF peak is shifted down by -2x, although a high frequency tail extends up to 25 MHz. FFP-TF linearization and skewing gives an optimal combination of low interference fringe frequency, efficient use of ADC time, and increased sensitivity through increased sample exposure time. 5.4 Optical Frequency Clock Optimization The OFC used in the 3D-OCT endomicroscopy system is a unique way to enable real-time imaging at speeds up to 100,000 axial lines per second by removing the requirement to sample a second interference signal and recalibrate the OCT data. Since FDML sweep linearization does not result in optimal PSF shapes as shown in Figure 5.9, the OFC must be used to remove residual v nonlinearity in the sweep to provide optimal axial resolution and minimal sidelobe levels. Although the OFC is extremely useful it presents additional performance tradeoffs not typically found in OCT imaging systems. These tradeoffs are a result of the link between the path imbalance in the OFC MZI used to generate the ADC sample clock and the imaging depth. First, the OFC introduces a tradeoff between maximum imaging depth and system noise. Second, it introduces an additional tradeoff between maximum imaging depth and axial resolution. These tradeoffs are described below. 5.4.1 Tradeoffs Between Imaging Depth and System Noise As described in Section 5.2, the OFC uses an unbalanced MZI and analog electronics to generate one ADC clock pulse every time the MZI interference signal crosses 0 V. These zero crossings are evenly spaced in optical frequency v so the OFC ensures that the OCT signal is sampled at evenly-spaced v intervals. One side effect of this setup is that the path mismatch in the MZI determines the maximum imaging depth of the system. As described in Chapter 1, the maximum imaging depth of a swept source OCT system is determined by the optical frequency spacing between consecutive samples in the interference signal. The OFC generates exactly two sample clock pulses for every MZI interference fringe period. Therefore Nyquist-limited sampling is achieved for OCT signal frequencies equal to the MZI fringe frequency and the MZI path imbalance is equal to the maximum imaging range. For sample ranges higher than the MZI path imbalance, the OCT signal would be under-sampled and aliasing would occur. This is prevented by placing a 100 MHz lowpass filter in the ADC circuitry. Since the system ADC operates at 200 MSamples/s, the MZI would ideally be set to produce 100 MHz interference fringes to obtain the largest possible imaging depth (4.7 mm in air for the 59 kHz linearized and skewed FDML laser). However, as shown in Figure 5.2, the SNR of the interference signal decreases at increased path imbalances. At 4.7 mm the fringe contrast decreases by > 18 dB compared to the optimal value at short imbalances. This results in a substantial increase in OFC sample clock jitter caused by detection of false zero crossings at the lower-amplitude start and end of the sweep. Clock jitter translates into erroneous assignment of v values prior to Fourier transformation, dramatically elevating the noise floor of the image. Loss of MZI SNR can also cause temporal walk-off of the OCT signal. The system requires a fixed number of sample points to be acquired during each axial line. If an insufficient number of clock events are generated due to OFC noise, a portion of the next sweep is stored as the previous sweep. This also causes artifacts in the resulting OCT image. In practice, the MZI imbalance was set to a less aggressive value to ensure the generation of low-noise images. MZI imbalances of 2.4 - 2.5 mm were typically used for human imaging studies. At this range, the MZI fringe contrast is decreased by less than 10 dB compared to the optimal value. No signal walkoff or noise floor fluctuations occurred at this setpoint. An imaging range of 2.4 mm in air corresponds to 1.7 mm in tissue, which is sufficient to ensure that penetration depth is limited by scattering and absorption in the tissue rather than the system parameters. 5.4.2 Tradeoffs Between Imaging Depth andAxial Resolution The MZI path imbalance, and therefore the maximum imaging depth, also affects the axial resolution of the 3D-OCT endomicroscopy system. This arises fundamentally from the finite coherence length of the FDML laser and the presence of chromatic dispersion in the long fiber cavity. As described in Chapter 2, the drive period of the FFP-TF in an FDML laser can be precisely synchronized to the optical roundtrip time of light in the cavity at only one wavelength. This wavelength is typically selected to be in the middle of the sweep. Chromatic dispersion causes a slight desynchronization between the FFP-TF and the light at the edges of the sweep. In addition to increasing cavity losses at those wavelengths, this also reduces the number of times the FFP-TF is effectively propagated through by the photons at the sweep edges. Reducing the number of FFP-TF transmission cycles results in a broadened instantaneous linewidth at the blue and red edges of the sweep, reducing the edge coherence lengths and thereby reducing the fringe contrast at large delays for these wavelengths. The tradeoff between imaging depth and axial resolution caused by these effects is illustrated in Figure 5.10 and Figure 5.10. When the MZI imbalance is set to 1.0 mm, the OFC MZI generates a fringe that covers all of the wavelengths contained in the FDML sweep. ADC clock events are therefore triggered over a time duration spanning the entire OCT signal (shown in magenta for a short imaging depth) and the entire tuning range of the laser is utilized. When the MZI imbalance is set to 4.5 mm, however, the imbalance exceeds the coherence length of the blue edge of the sweep. The MZI fringe duration decreases and no clock events are generated during the last 25% of the sweep. Therefore the OCT signal is not entirely sampled and 25% of the tuning range is unused. This decreases the spectral content of the sampled OCT signal and reduces axial resolution by a corresponding 25%. In practice, as discussed in Section 5.4.1, the MZI path imbalance was set to 2.4 - 2.5 mm. At this distance the axial resolution decreases by only -10% due to MZI fringe dropout. Further degradation of the axial resolution is caused by jitter in the OFC that tends to broaden the PSFs by sampling the OCT signal at unevenly spaced optical frequency positions. The axial resolution of the system was measured as described in Section 5.5.1. 5.5 System Characterization The specialized OFC, ADC, and signal processing software used in the 3D-OCT endomicroscopy system significantly affect the overall performance. Axial resolution, sensitivity and sensitivity rolloff of the overall system are significantly different than the values supported by the FDML laser itself. The true resolvable feature size or "true resolution" of the system is also a function of the spatial sampling density in addition to the optical resolution determined by the laser tuning range and imaging probe spot size. These performance metrics were characterized for the overall system as described below. 5.5.1 Axial Resolution Axial resolution was measured by replacing the imaging probe with an equivalent length of SMF-28 fiber, bulk collimating lens, neutral density filter, and metallic mirror. The mirror provided a point reflection at an adjustable position away from the fiber tip while the filter provided 20 dB of single-pass attenuation to prevent detector saturation. The OFC MZI was set to a path mismatch of 2.45 mm corresponding to the maximum imaging range typically used for clinical studies. The resulting OCT interference signal was captured by the ADC and converted to a linear PSF in real time using the system's signal processing software. The linear PSF at an imaging depth of 70 gm is shown in Figure 5.12. The FWHM of the linear PSF is 8.0 tm in air compared to 6.4 pm in air as measured with the FDML laser alone. The corresponding resolution in tissue is 5.8 pm for the complete system compared to 4.6 [pm for the laser alone. As discussed above, the degradation is caused partially by OFC clock dropout at the blue edge of the MZI fringe and partially by OFC jitter. 5.5.2 Sensitivity Rolloff System sensitivity was measured using the same sample arm setup as described in Section 5.5.1. Sensitivity is obtained by comparing the peak signal value at a short imaging depth to the system noise floor with the sample arm blocked, correcting for the attenuation of the neutral density filter. With 15 mW of power (backward sweep only) incident on the sample a sensitivity of -113 dB was recorded. The theoretical shot-noise limited sensitivity for a backward sweep duration of 9.9 ls, 15 mW of power on the sample, and 50% detector sensitivity is -117 dB from Equation 1.8. Single-pass losses in the sample arm and Michelson interferometer of -2 dB account for the variation in sensitivity away from the shot-noise limited value. Sensitivity rolloff for the complete system was measured by translating the mirror away from the fiber tip and recording a PSF at each new position. The results of this measurement are shown in Figure 5.13. Compared to the rolloff induced by the laser itself, the system rolloff is slightly more severe. The 6 db rolloff position is at 1.2 mm for the complete system compared to 1.6 mm for the laser only. This increase in rolloff is also due to OFC clock dropout and jitter. 5.5.3 True Spatial Resolution When characterizing the resolution of a 3D-OCT system it is important to differentiate between optical resolution, spatial sampling density, and true spatial resolution. Optical resolution, determined by the focal spot size of the sample optics and the width of the axial point spread function, defines the best resolution that the system can theoretically obtain. Spatial sampling density is the distance between consecutive axial scans in X and between consecutive frames in Y. Spatial sampling density in Z can be set arbitrarily by selecting the Fourier transform length in a swept source setup. True resolution is determined by a combination of optical resolution and spatial sampling density, and defines the smallest feature size that can actually be visualized. It should be noted that optical resolution is not constant with imaging depth, since the beam diverges away from the focal position. In addition, the ability to resolve structures depends on feature contrast. The Nyquist criterion requires at least two spatial samples in each dimension for every optical resolution element. For the OCT system reported here the optical resolution in tissue is 12 x 12 x 6 lIm, so spatial samples should be acquired every 6 x 6 x 3 htm. The XY spatial sampling density is fundamentally limited by imaging speed, tissue surface area, and maximum imaging duration. Since surface area and imaging duration are determined by anatomy and physiology, the only parameter available for increasing spatial sampling density is imaging speed. Therefore high imaging speeds, as determined by the laser sweep rate and data acquisition capacity, are required to obtain true microscopic 3D-OCT resolution. The XY spatial sampling density of the 3D-OCT endomicroscopy system constructed here could be adjusted by varying the probe rotation rate and pullback speed. To obtain a reasonable tradeoff between spatial sampling density and field of view, the rotational speed was typically set to 60 Hz and the pullback speed to 1.0 mm/s. With axial image lines acquired at 59 kHz, this gave 983 lines per frame with 16.7 jim between consecutive frames. The transverse pixel spacing was approximately 8 jim, given an average imaging depth of 1240 jLm in tissue (70% of the maximum imaging depth) and a corresponding circumferential scan length of 7.8 mm. The actual transverse pixel spacing varies with depth due to the rotational nature of the probe. Axial pixel spacing was 3.5 jm with a 1024 point Fourier transform performed on 864 k-space samples per line. These parameters gave a true spatial resolution of 16 x 33 x 7 3 jtm (3696 jim3) compared to an optical resolution of 12 x 12 x 6 jm (864 jim ). In the future, faster imaging speeds and increased probe rotational rates will be necessary to make full use of the optical resolution supported by the FDML laser and imaging probes. The current performance nevertheless represents a significant improvement over previous endoscopic swept source OCT systems, which used polygon mirror based swept lasers and achieved optical resolutions of 15 x 15 x 7 jtm with XY spatial sampling densities of only 25 x 33 jim due to restricted imaging speeds of 10 kHz [9, 15]. 3 Consequently, the true resolution of these previous systems was 50 x 66 x 7 gm (23,100 gim ) at imaging speeds of 10 kHz in the esophagus and 54 kHz in the coronary artery. 5.6 Data Post-Processing Several data post-processing algorithms were developed in order to achieve optimal visualization of tissue microstructure. Since the mucosal structure of the GI tract is difficult to evaluate in a cylindrical form, the individual 3D-OCT frames must be converted from radial images to rectangular image. To facilitate the formation of enface images at arbitrary depths, the frames must then be flattened to prevent curvature artifacts. Finally, the frames must be stored as compressed JPEG files to prevent memory overflow in the commercial 3D rendering software used to analyze the datasets. These steps are described in Sections 5.6.1 and 5.6.2 below. 5.6.1 FrameFlattening The frame flattening algorithm is shown in Figure 5.14. Rectangular blocks indicate automated steps while trapezoidal blocks indicate steps requiring user input. Diamond-shaped blocks are decision steps and cylindrical blocks are data operations. The flattening algorithm works by detecting the imaging probe sheath's irregular surface and then shifting each axial line of each frame such that the sheath becomes flat. Since the sheath is nominally in direct contact with the tissue surface, this simultaneously flattens the tissue as well. 100 The algorithm consists of three main sections as shown by the three main columns in the algorithm flowchart. First, the software obtains user input for several processing parameters. Since each frame is stored as a series of axial lines in the LightLab software (as opposed to a radial image), no processing is required to convert the radial frame to a rectangular frame. Each frame is stored as a series of Fourier transformed logarithmic image lines with 12 bit resolution. Three representative test frames are parsed from the beginning, middle, and end of the 3D dataset in order to obtain the required user input. The frames are displayed as rectangular Matlab figures. The. user is first asked to set an axial exclusion range to remove fixed line artifacts from the image. These artifacts are typically the result of a small reflection from an optical surface in the proximal section of the probe and appear as moderately bright lines inside the probe lumen. All pixels located inside the exclusion range are set to zero value to avoid erroneous edge detection. Next the user is asked to input a horizontal wrapping point for the image. The GI tissue is not typically in contact with the probe over the entire 360' rotation of a single frame. The horizontal wrapping point sets the position for the first axial line of the rectangular dataset and allows the user to centre the useful portion of the rotation in the middle of the frame. Next the user is given the option to set up small rectangular masks to remove any residual artifacts not covered by the axial exclusion range that may corrupt the sheath detection. At this point the algorithm applies a 6 x 3 pixel Gaussian smoothing filter to remove speckle noise in the three representative test frames. The user is shown each filtered frame and asked to provide a noise threshold level such that the sheath edge reflection remains above the threshold but all structures at more shallow axial pixels are set to zero. This is the final piece of manual input required for the algorithm. Next the post-processing software sequentially reads in each frame and attempts to detect the sheath boundary. This is done by removing the axial exclusion zone and masks, horizontally shifting the frame according to the user-defined wrapping point, applying the Gaussian filter, and thresholding out pixel values below the user-defined noise level. Then a Sobel edge detector is applied to the frame to generate a vector of edge points for each axial line. Sobel edge detection is a rapid, robust method for locating edges based on 2D image information. The sheath is detected by isolating the first edge point in each line. For lines where no edge is detected, the missing sheath location is estimated by interpolating between neighbouring lines. Finally a 12 th order polynomial fit is applied to the sheath edge in order to smooth out abrupt intra-frame jumps caused by the Sobel algorithm. After the sheath location has been stored for each frame, a 10-frame rolling average is applied to smooth out inter-frame jumps. 101 5.6.2 JPEG Conversion The final section of the algorithm shifts each axial line in each frame and saves the output as a stack of JPEG files. Each frame is sequentially read in a second time. The stored sheath location for the frame is recalled and each axial line is axially shifted such that the sheath, and therefore the top surface of the tissue, becomes horizontally oriented. No thresholding or Gaussian filtering is applied at this stage in order to preserve the original image data. A Matlab figure with the correct number of pixels is created and saved as a high-quality 3-channel 24 bit JPEG with each channel encoding an 8-bit colour level. Once this process is completed for each frame the 3D-OCT data file is closed and the program terminates. In order to visualize the 3D-OCT data using a commercial rendering package, the high-quality 24-bit JPEGs created by the flattening algorithm must be compressed to lower-quality versions. The rendering package used throughout this thesis work (ResolveRT, Mercury Computer Systems) is not capable of manipulating - 1000 x 512 x 1200 datasets when each pixel is a 24-bit value. A second JPEG stack is therefore created that consists of compressed 8-bit data. The greyscale image quality is very similar to the original 24-bit data since the human eye has difficulty distinguishing more than 256 levels of grey. 5.7 Image Visualization 3D-OCT imaging enables the employment of a variety of powerful visualization methods in order to comprehensively assess tissue microstructure. Full 3D renderings of the dataset can be created to inspect the tissue at any spatial orientation. Individual 2D image planes can also be defined, enabling arbitrary cross-sectional analysis of the tissue. For example, en face sections can be produced to provide a virtual microscopic view of the tissue at any depth. Cross-sectional images can then be generated with exact registration to en face features and anatomic landmarks. Additionally, consecutive 2D images from a densely sampled 3D dataset can be averaged to reduce speckle noise and improve the visualization of subtle tissue structures. Quantitative 3D measurements can also be performed on tissue structures in order to assess pathologic state. In general, 3D-OCT imaging provides a more complete understanding of the tissue architectural morphology compared to 2D imaging. The following sections describe several visualization methods developed as part of this thesis work. Preliminary data was acquired in the rabbit colon using a modified commercial cardiovascular imaging probe. 5.7.1 3D Rendering 3D rendering enables inspection of the imaged tissue from arbitrary orientations and is also helpful for orienting individual 2D image planes within the dataset. Figure 5.15 illustrates the generation of 3D-OCT renderings from a stack of 885 radial frames. Figure 5.15(A) shows a single radial frame acquired in the rabbit colon, while Figure 5.15 (B) shows a cutaway view of the complete rendered dataset. No unfolding 102 or frame flattening was applied while generating the rendering shown in Figure 5.15 (B) so the data retains the cylindrical shape of the original acquisition. The cutaway view enables visualization of the luminal surface, with crypts appearing as dark features surrounded by bright white bands of lamina propria. The enlarged view of the epithelium shown in Figure 5.15 (C) illustrates crypt structure in a single frame. Figure 5.15 (D) shows a 3D rendering formed after unfolding and flattening the dataset as described in Section 5.6. The unfolded visualization enables clearer appreciation of tissue morphology and is more amenable to comparisons with endoscopic images and biopsy-based histology. 5.7.2 OrthoplaneSectioning After application of the unfolding and flattening algorithm, cross-sectional OCT images with arbitrary orientations can be generated that are precisely registered to the surface of the tissue. These orthoplanes can be scanned through the tissue volume at any location, facilitating detailed inspection of tissue microstructure. Figure 5.16 illustrates the concept of orthoplane analysis for the rabbit colon dataset shown in Figure 5.15(D). Figure 5.16(A) shows an en face XZ orthoplane image located in the epithelial tissue layer of the colonic mucosa. The mottled appearance of the image is a result of the colonic crypts in the epithelium. Orthogonal cross-sectional images showing depth-resolved structure, such as the YZ image shown in Figure 5.16(B), can be aligned to surface features for enhanced analysis of tissue microstructure. The layered nature of the mucosa can be clearly appreciated in the YZ cross-sectional orthoplane. 5.7.3 Projection Viewing Since the 3D-OCT data set is sampled with a high spatial density, consecutive cross-sectional or en face slices can be averaged or "projected" to reduce speckle noise without significantly blurring image features. This concept is illustrated in Figure 5.17. The image in Figure 5.17(A) was formed by calculating the mean of 7 consecutive YZ slices, equivalent to averaging the data over a 21 jtm thick section. A 21 [tm section is smaller than the dimension of two epithelial cells in the rabbit colon, so tissue microstructure is largely constant and minimal image blurring is observed compared to the single image in Figure 5.16(B). Since the speckle size is approximately equal to the 9 jtm focal spot size of the probe used to collect this data, the speckle pattern is decorrelated over the 21 jtm section. Averaging a decorrelated speckle pattern provides significant rejection of speckle noise and enhances tissue contrast. Figure 5.17(B) shows a 3x enlarged view of a region of Figure 5.17(A). The 3D-OCT image correlates well with representative histology of colonic tissue from the same animal (Figure 5.17(C)), although the mucosal surface is flattened in the 3D-OCT images due to compression by the imaging probe. Variations in apparent layer thickness are due to probe compression as well as tissue shrinkage that occurs during 103 histology processing. The dark vertical bands in the 3D-OCT image are caused by fecal material on the luminal surface. 5.7.4 Linear En FaceImages 3D-OCT data is typically stored and visualized using a logarithmic scale in order to display the full dynamic range (> 50 dB) of the data acquired over a typical tissue depth of -1000 [m. Logarithmic compression of the data, however, has the unwanted side-effect of reducing the effective transverse resolution in enface images. This occurs because the system's transverse point spread function, defined by the width of the incident OCT beam, is broadened when a logarithmic transform is applied. This makes small transverse features appear wider than their true size. A comparison between linear and logarithmically scaled en face images is shown in Figure 5.18. Both images were generated using the same dataset by averaging the data axially over a 20 tm range centered at a tissue depth of 400 tm. In the linear image shown in Figure 5.18(A) the transverse crypt features appear sharper and more defined than in the logarithmic image shown in Figure 5.18(B). Due to the lower dynamic range of the linear image, however, colour map saturation artifacts are clearly visible at the right hand side of the linear image (red arrows). Conversely, the logarithmic image is able to display the wider dynamic range of the 3D-OCT data and does not produce equivalent artifacts. For most GI imaging applications a logarithmic pixel compression is desirable to allow visualization of the complete range of pixel intensities found in highly scattering biological media. 104 5.8 Figures Sample RM 1C 2 3 E N time Av Av TRG +DA- +DA- OCT OFC OFC OCT Custom 200 MHz DAQ / DSP System Personal Computer Figure 5.1. 3D-OCT endomicroscopy system schematic. C, circulator. PIU, patient interface unit. RM, reference mirror. MZI, Mach-Zehnder interferometer. P, photodetector. DA, differential amplifier. MZI, Mach-Zehnder input. TRIG, sweep trigger input. OFC, optical frequency clock input. OCT, OCT signal input. RAM, random access memory. FFT, fast Fourier transform. Left inset shows the principle of OFC generation using the MZI output. Zero crossings are unevenly spaced in time, but evenly spaced in optical frequency v.Originally published in [16]. PSF Rolloff: 60 kHz FDML Laser 0 -FDML laser used in 3D-OCT Figure 5.2. endomicroscopy system. FFP-TF, fiber Fabry-Perot tunable filter. ISO, isolator. SOA, semiconductor optical amplifier. 105 3 2 Imaging Range [mm] 1 4 Figure 5.3. PSF rolloff measured using a 60 kHz FDML laser and an unbalanced MZI. 6 dB rolloff point is at 1.6 mm. Originally published in [16]. FFP-TF Amplitude Response Linear PSF: 60 kHz FDML Laser 20 E *--15- ,j ir ,.10 0.04 0.1 0.08 0.06 Imaging Depth [mm] 04 0.12 60 Applied FFP Drive Wave FFP Drive Waveform for Linear Sweep 1.5 1 Time 05 180 160 Figure 5.5. Amplitude response of an FFP-TF as the drive frequency is varied over three regions. Clear harmonic peaks are observed near 56 kHz and 163 kHz. Figure 5.4. Linear PSF measured with 60 kHz FDML laser at a short imaging depth. Axial resolution is 6.4 um in air or 4.6 um in tissue. -1' 0 80 100 120 140 Frequency [kHz] I x 10 Time x104 Figure 5.7. Actual drive wave applied to FFP-TF after correcting for amplitude and phase response of filter. Figure 5.6. Desired FFP-TF response for linear, skewed sweep in the backward direction. FFP-TF Amplitude Response FDML Output Using Skewed Linear Drive Wave 2- 240 60% Backward 0% Forward Sweep I %4 ~ Sweep E C 2M6 1200 MI C q .0 E 00 no -14 -12 -10 -8 -6 -4 -2 0 2 Time [us] Figure 5.8. Amplitude and phase of interference signals generated by FDML laser and unbalanced MZI after application of linearized and skewed FFP-TF drive wave. 106 C 5 10 30 25 20 15 Frequency [MHz] 35 40 Figure 5.9. PSFs generated at the same imaging depth for various FFP-TF drive waveforms. Ideal PSF is obtained with postprocessing recalibration. Other PSF's were not recalibrated. Figure 5.10. U-;u -igure 5.11. signal durations wnen MLI t-L; signal aurations wnen MLI imbalance is set to 4.5 mm. Clock events are generated for only a portion of the sweep. 25% of the OCT signal is not sampled, reducing the axial resolution. imbalance is set to 1.0 mm. Clock events are generated for the entire sweep duration (red bars) and the entire OCT signal can be sampled. PSF Rolloff: Complete System Linear PSF: Complete System 1 ,0.8 EA 0.6 0.4 E __ .4 80 60 Imaging Depth [um] 1.5 1 0.5 Imaging Range [mm] 120 Figure 5.13. PSF rolloff measured with complete 3D-OCT endomicroscopy system. 6 dB rolloff point is at 1.2 mm. Figure 5.12. Linear PSF measured with complete 3D-OCT endomicroscopy system at a short imaging depth. Axial resolution is 8.0 um in air or 5.8 um in tissue. 107 Figure 5.14. Flowchart of 3D-OCT data post-processing algorithm for frame flattening and JPEG conversion. Frames are flattened by detecting the probe sheath and shifting each axial line such that the sheath and tissue surface become horizontally oriented. 108 Figure 5.15. Construction of 3D-OCT renderings by stacking a series of radial frames. A: Single radial frame of rabbit colon. B: Cutaway view of rendering formed by stacking 885 frames. C: Enlarged view of epithelium in A. D: 3D rendering of unfolded dataset after frame flattening. Originally published in [16]. 109 Figure 5.16. Single XZ and YZ orthoplanes near the middle of the dataset shown in Figure 4.15. A: En face XZ orthoplane showing details of colonic epithelium in the rabbit. Crypt structures are visible between bright bands of lamina propria. Red arrow pairs indicate location of cross-sectional image in B. B: Longitudinal YZ orthoplane near the centre of the tissue volume. The layered mucosal structure can be appreciated over a range > 1000 um in tissue. Red arrow pairs indicate location of en face image in A. AC LI M SM MC MP ML SE AC: Absorbing Cells CM: Colonic Mucosa LP: Lamina Propria MM: Muscularis Mucosa SM: Submucosa MC: Circular Muscle MP: Myenteric Plexus ML: Longitudinal Muscle SE: Serosa FT: Fatty Tissue LP MM SM MP ML Figure 5.17. Averaging or "projecting" images over a thin section reduces speckle noise and improves contrast. A: Longitudinal YZ image formed by averaging 7 consecutive cross-sectional images over a span of 21 um. Speckle is reduced compared to the single image shown in Figure 5.16(B). B: Enlarged view of A showing all colonic layers. C: comparison to histology from excisional biopsy specimen. Originally published in [16]. 110 Figure 5.18. Comparison of linear and logarithmically scaled en face images of the human colon. A: Linear scaled en face image at a tissue depth of 400 um averaged over a range of 20 um. Saturation artifacts are visible (red arrows). B: Logarithmically scaled en face image at the same location, also averaged over a range of 20 um. Features appear sharper in the linear image but saturation artifacts appear due to insufficient dynamic range. 111 5.9 References [1] W. Drexler, U. Morgner, R. K. Ghanta, F. X. Kirtner, J. S. Schuman, and J. G. Fujimoto, "Ultrahigh-resolution ophthalmic optical coherence tomography," Nature Medicine, vol. 7, pp. 502-507, Apr 2001. [2] G. J. Tearney, M. E. Brezinski, S. A. Boppart, B. E. Bouma, N. Weissman, J. F. Southern, E. A. Swanson, and J. G. Fujimoto, "Catheter-based optical imaging of a human coronary artery," Circulation,vol. 94, p. 3013, 1996. [3] B. E. Bouma, G. J. Tearney, C. C. Compton, and N. S. Nishioka, "High-resolution imaging of the human esophagus and stomach in vivo using optical coherence tomography," Gastrointestinal endoscopy, vol. 51(4) Pt 1, pp. 467-74, Apr 2000. [4] G. J. Tearney, M. E. Brezinski, J. F. Southern, B. E. Bouma, S. A. Boppart, and J. G. Fujimoto, "Optical biopsy in human urologic tissue using optical coherence tomography," The Journalof urology, vol. 157, pp. 1915-9, May 1997. [5] P. F. Escobar, J. L. Belinson, A. White, N. M. Shakhova, F. I. Feldchtein, M. V. Kareta, and N. D. Gladkova, "Diagnostic efficacy of optical coherence tomography in the management of preinvasive and invasive cancer of uterine cervix and vulva," International Journal of Gynecological Cancer,vol. 14, pp. 470-474, MAY-JUN 2004. [6] A. R. Tumlinson, B. Povazay, L. P. Hariri, J. McNally, A. Unterhuber, B. Hermann, H. Sattmann, W. Drexler, and J. K. Barton, "In vivo ultrahigh-resolution optical coherence tomography of mouse colon with an achromatized endoscope," Journal of Biomedical Optics, vol. 11, pp. -, Nov-Dec 2006. [7] P. R. Herz, Y. Chen, A. D. Aguirre, J. G. Fujimoto, H. Mashimo, J. Schmitt, A. Koski, J. Goodnow, and C. Petersen, "Ultrahigh resolution optical biopsy with endoscopic optical coherence tomography," Optics Express, vol. 12, pp. 3532-3542, JUL 26 2004. [8] A. R. Tumlinson, J. K. Barton, B. Povazay, H. Sattman, A. Unterhuber, R. A. Leitgeb, and W. Drexler, "Endoscope-tip interferometer for ultrahigh resolution frequency domain optical coherence tomography in mouse colon," Optics Express, vol. 14, pp. 1878-1887, Mar 6 2006. [9] S. H. Yun, G. J. Tearney, B. J. Vakoc, M. Shishkov, W. Y. Oh, A. E. Desjardins, M. J. Suter, R. C. Chan, J. A. Evans, I. K. Jang, N. S. Nishioka, J. F. de Boer, and B. E. Bouma, "Comprehensive volumetric optical microscopy in vivo," Nature Medicine, vol. 12, pp. 1429-1433, Dec 2006. [10] T. Takayama, S. Katsuki, Y. Takahashi, M. Ohi, S. Nojiri, S. Sakamaki, J. Kato, K. Kogawa, H. Miyake, and Y. Niitsu, "Aberrant crypt foci of the colon as precursors of adenoma and cancer," New EnglandJournalof Medicine, vol. 339, pp. 1277-1284, Oct 29 1998. 112 [1 l] E. Brinkmeyer and R. Ulrich, "High-Resolution Ocdr in Dispersive Wave-Guides," Electronics Letters, vol. 26, pp. 413-414, Mar 15 1990. [12] M. Wojtkowski, R. Leitgeb, A. Kowalczyk, T. Bajraszewski, and A. F. Fercher, "In vivo human retinal imaging by Fourier domain optical coherence tomography," Journalof Biomedical Optics, vol. 7, pp. 457-463, 2002/07/ 2002. [13] S. H. Yun, G. J. Tearney, J. F. de Boer, N. Iftimia, and B. E. Bouma, "High-speed optical frequency-domain imaging," Optics Express, vol. 11, pp. 2953-2963, Nov 3 2003. [14] C. M. Eigenwillig, B. R. Biedermann, G. Palte, and R. Huber, "K-space linear Fourier domain mode locked laser and applications for optical coherence tomography," Optics Express, vol. 16, pp. 8916-8937, Jun 9 2008. [15] B. J. Vakoc, M. Shishko, S. Yun, W. Y. Oh, M. J. Suter, A. E. Desjardins, J. A. Evans, N. S. Nishioka, G. J. Tearney, and B. E. Bouma, "Comprehensive esophageal microscopy by using optical frequency-domain imaging," GastrointestinalEndoscopy, vol. 65, pp. 898-905, May 2007 2007. [16] D. C. Adler, Y. Chen, R. Huber, J. Schmitt, J. Connolly, and J. G. Fujimoto, "Three-dimensional endomicroscopy using optical coherence tomography," Nature Photonics, vol. 1, pp. 709-716, Dec 2007. 113 CHAPTER 6 6.0 High Speed Imaging Probes for Clinical 3D-OCT 114 6.1 Motivation Imaging probes are a critical component of 3D-OCT systems, with probe performance directly impacting overall image quality and clinical utility. Gastrointestinal (GI) imaging with 3D-OCT requires beam delivery through a probe that is flexible, robust, and small enough to fit through the working channel of a conventional endoscope. Probe design can be optimized for different imaging requirements and entails tradeoffs between the scan pattern, scan speed, transverse resolution, working distance, size, robustness, and cost. Initial animal studies and validation of the 3D-OCT endomicroscopy system described in Chapter 5 were performed using a modified commercial cardiovascular probe supplied by LightLab Imaging Inc. This commercial probe was a proximally-actuated radial imaging device that combined rapid rotary motion (up to 80 Hz) with a linear pullback to image a cylindrical volume. The commercial probe was helpful during development of the 3D-OCT system and image processing methods, but its modified cardiovascular design was not robust or stable enough for GI applications. The diameter of the commercial probe was minimized to allow access to narrow blood vessels, which reduced mechanical strength and rendered the probe vulnerable to damage during GI endoscopy. Furthermore, the small diameter torque coil used to transfer rotational motion to the distal tip was somewhat unstable when the probe was passed through anatomic bends in the colon. Finally, the maximum insertion length of the probe was insufficient for studying the transverse colon and right colon. For these reasons an improved spiral-scanning probe that was optimized for GI imaging was developed in this portion of the thesis work. Two different optical lens systems were developed in parallel to reduce the risk of failure. One lens system used a series of fused optical fibers to create an integrated lens with minimal diameter and backreflected power. The second lens system used a bulk graded index (GRIN) lens to achieve high transmission, efficient back-coupling, and good beam quality. Both lens systems were designed to be compatible with a common torque coil for translation of rotational motion, plastic sheath for electrical and biological isolation, and proximal fiber connector for attachment to the patient interface unit (PIU) developed by LightLab Imaging Inc. The bulk micro-optic GRIN design was found to provide better performance and much higher assembly yields than the fuser fiber design, although the fused fiber design could be used in the future for applications such as needle-based imaging requiring the smallest possible probe diameter. 6.2 Fused Fiber Lens Systems Fused fiber lenses consisting of single mode fiber, non-waveguiding "coreless" fiber, and graded index multimode fiber have been described previously for OCT imaging probes [1-4]. In a fused fiber probe the light travels from the proximal end to the distal tip through an arbitrary length (typically several meters) 115 of single mode fiber such as SMF-28. A short section (typically several hundred microns) of coreless fiber is used to expand the 9.2 jtm beam diameter at the SMF-28 core. The expanded beam then enters a graded index multimode fiber section that focuses the beam to a desired radius and working distance. Each fiber section is joined together using fusion splicing to ensure excellent optical contact between the elements, thereby minimizing backreflected light that can corrupt OCT images. A variety of polishes can be applied to the lens tip in order to change the direction of the beam or further reduce backreflection. For this section of the thesis work two different fused fiber lens systems were developed based on an initial design by LightLab Imaging Inc. One design used standard commercially-available multimode fiber for the focusing element, while the other used custom multimode fiber with a low index gradient. The low gradient fiber produces a weaker focusing effect, allowing the working distance to be extended compared to the standard multimode fiber. In both cases, standard commercially-available coreless fiber was used for beam expansion. Both lenses were designed using ZEMAX modeling software as described in Section 6.2.1. A performance comparison between the two designs is provided in Section 0 6.2.1 ZEMAX Modeling - StandardMultimode Fiber The ZEMAX model of the fused fiber lens system using commercial multimode fiber is shown in Figure 6.1. The lens is formed by fusing a 145 [tm length of multimode fiber between 493 tm and 230 gim lengths of coreless fiber. The second length of coreless fiber is used to provide space for end tip polishing as described in Section 6.2.3. The working distance of the lens was set to 1540 im in water. This working distance was required in order to propagate the beam outside of the probe body (-1040 jtm radius, filled with water) and 500 gm into the tissue. The refractive index properties of the multimode fiber was obtained from the supplier (Newport Inc.) and the coreless fiber was modeled as pure silica with a diameter of 125 jim. The multimode section had a core diameter of 100 jtm and was used as the field stop in the ZEMAX model. Optimization of the design was performed by setting the three fiber lengths as independent variables. The optimization function was set up to place a beam waist at the output plane, effectively setting the focus at the final surface in the model. Equal weight was placed on the X and Y components of the beam for optimization. A Gaussian skew beam with a diameter of 9.2 gm was launched from the first surface of the model to emulate the mode field of SMF-28 fiber, and Gaussian modeling was used for the optimization loop. Optimization resulted in the component lengths shown in Figure 6.1(A). Figure 6.1(B) shows the spot diagram at the output surface, giving an estimated beam radius of 15 jm. This spot size is - 2x larger than ideally desired for imaging small features such as crypts or Barrett's esophagus glands in the human GI tract, but would provide reasonable performance for many applications. 116 6.2.2 ZEMAX Modeling - Custom Multimode Fiber The ZEMAX model of the fused fiber lens system using custom multimode fiber is shown in Figure 6.2. The lens is formed by fusing a 559 jtm length of multimode fiber between 573 [tm and 200 jim lengths of coreless fiber. The second length of coreless fiber is used to provide space for end tip polishing as described in Section 6.2.3. As with the previous design, the working distance of the lens was set to 1540 jim in water. The refractive index properties of the multimode fiber was obtained from the supplier (LightLab Inc.) and the coreless fiber was modeled as pure silica with a diameter of 125 jtm. The multimode section had a core diameter of 62.5 jtm and was used as the field stop in the ZEMAX model. Some vignetting of the expanded beam is visible at the first surface of the multimode fiber section in Figure 6.2. This resulted in a slight decrease in transmission but gave an improved beam radius at the output surface. Optimization of this design was performed in the same manner as the previous design. Figure 6.2(B) shows the spot diagram at the output surface, giving an estimated beam radius of 8 Rm. This spot size and working distance is close to ideal for GI imaging applications. 6.2.3 Polishing In order to produce a radial scan pattern when the lens is rotated, the beam must exit the lens perpendicular to the long axis of the fiber. This is accomplished by placing a 400 angle polish on the front face of the lens tip to tilt the beam by 80' as shown in Figure 6.3. A small air pocket at the tip of the lens is formed by adding an ultrathin plastic sheath to the lens and filling the end with epoxy. This air pocket creates total internal reflection when the light strikes the interface between the coreless fiber and the air, tilting the beam by 80' relative to it's original direction as shown by the red arrow in Figure 6.3. A 900 tilt would result in the beam striking the probe sheath at normal incidence, giving a high backreflection that can corrupt OCT images. A second 50 polish is also required on the longitudinal face of the fiber lens as shown in Figure 6.3. This polish removes the radius of the fiber and prevents strong negative lensing as the beam exits the coreless fiber and enters the water inside the probe lumen. Polishing was conducted using a mechanical disc polisher and a purpose-built jig to hold the fiber in place. The fiber was placed in a standard single mode fiber chuck and the chuck was clamped into the jig. The 500 face angle polish was performed first, then the entire jig was rotated 450 and the longitudinal polish was applied. Polishing was done in stages down to a final grit size of 0.3 jim. 6.2.4 FabricationTolerances One significant drawback of the fused fiber lens designs is that they are highly intolerant to length errors in the multimode fiber section. Table 6.1 shows ZEMAX model results of tolerance testing for both the commercial and custom multimode fiber lenses. In these simulations the length of the multimode 117 section was varied by -30 tim to +50 jtm in steps of 10 jtm and the resulting change in working distance was recorded. With a desired focal depth of 500 jtm in tissue the maximum variation in working distance that can be tolerated is -400 jtm / +500 jim. At negative variations greater than -400 jtm the focus would be set inside of the probe's outer sheath, and at positive variations greater than +500 jtm the focus would be too deep to generate sufficiently high resolution of epithelial tissue layers. Table 6.1 shows that the commercial multimode fiber reaches these tolerance limits at length variations of +/- 20 jm. When the length of the commercial multimode fiber was reduced by more than 20 jam the focus shifted to inside of the last coreless fiber section. The custom multimode fiber reaches the defined tolerance limits at even smaller length variations of < +/- 10 jam. Since each fiber section must be cleaved and fusion spliced manually, this resulted in a very low assembly yield for both types of fused fiber lenses. Poor assembly yield was the main reason that the fused fiber probes were not pursued further, although they do provide exceptionally small diameters and may be useful for certain imaging applications. Commercial: Multimode Length Change in WD [umrn Error [um] N/A -30 -278 -20 -397 -10 0 0 274 10 473 20 614 30 719 40 799 50 Custom: Change in WD [um] 2344 3830 2886 0 -578 -799 -920 -996 -1049 Table 6.1. Variation in lens working distance (WD) resulting from length errors in commercial and custom multimode fiber lengths in fused fiber probes. 6.2.5 MeasuredPerformance The transmission, back-coupling, and working distance were measured for one of each type of fused fiber lens representing the best fabricated units. Spot sizes were estimated by entering the measured working distance back into the ZEMAX models and allowing the fiber lengths to vary to produce a focal spot at that location. The spot size was then recorded from the model output. This was necessary since spot size could not be measured directly using diagnostic equipment at hand in the laboratory. The optimal fused fiber lens incorporating commercial multimode fiber produced a transmission of 71%, back-coupling efficiency of 11%, and working distance of 203 jm giving an estimated spot size (diameter) of 4.3 jim. The optimal lens incorporating custom multimode fiber produced a transmission of 118 96%, back-coupling efficiency of 75%, and working distance of 508 [tm giving an estimated spot size (diameter) of 5.4 Rm. Several other fused fiber lenses produced vastly different working distances well outside the desired range for GI applications. While the small size of the fiber probes is highly desirable, automated splicing and cleaving equipment would be required to reproducibly construct the devices. 6.3 Micro-Optic Lens Systems Bulk GRIN lenses are used for a wide variety of applications in photonics and biomedical optics due to their small size and high optical quality. A bulk micro-optic lens system consisting of a GRIN lens, fiber pigtailed glass ferrule, metallized angle prism and epoxy spacer was designed as an alternative to the fused fiber lenses described in Section 6.2. The micro-optic lens was found to provide good performance and, importantly, much higher assembly yields compared to the fused fiber lenses. Figure 6.4 shows a schematic diagram of the micro-optic lens system and the complete distal probe tip. A length of SMF-28 fiber running to the proximal end of the probe is terminated in a glass ferrule. The beam exiting the fiber expands in an epoxy gap that is transmissive at 1310 nm. This epoxy gap serves the same function as the coreless fiber in the fused lens design, allowing the input aperture of the bulk GRIN lens to be filled more effectively. The GRIN lens is a commercially available element with a numerical aperture (NA) of 0.46 and a pitch of 0.25 at 1310 nm. An angle prism with a metallized hypotenuse is glued to the GRIN lens to direct the beam out the side of the probe, serving the same function as the angle polish in the fused lens design. The second surface of the GRIN lens is polished to 50 to ensure that the beam exits the probe 100 off of normal incidence to the sheath. All of the micro-optic components are glued to a stainless steel hypo tube using rigid epoxy. The lens system is joined to the flexible torque coil using an outer hypo tube that is epoxied to both the lens tube and the torque coil. A window is ground in the outer hypo tube to allow light to exit the lens. A transparent plastic sheath encapsulated the entire probe, providing electrical and biological isolation. Water is flushed down the probe from the proximal end to improve index matching between the angle prism, sheath, and tissue and also to provide lubrication for rotational motion. 6.3.1 ZEMAX Modeling A complete ZEMAX model of the micro-optic probe, including the sheath, water and tissue, was constructed to optimize the epoxy gap size. The ZEMAX model is shown in Figure 6.5. The model is constructed as a folded mirror image to allow estimation of the back-coupling efficiency. The sheath is modeled as two surfaces with one-dimensional curves corresponding to the inner and outer radius of the sheath. An 8' angle is introduced at the second surface to account for the angle polish on the glass ferrule. Note that the ray trace used to generate Figure 6.5(A) begins inside the ferrule, whereas the Gaussian 119 beams used for optimization and spot size estimation were launched from the output surface of the ferrule. After selection of an appropriate GRIN lens from the catalogue of a commercial supplier, the only free parameter in this model was the length of the epoxy spacer. This was optimized by constructing a merit function that placed a beam waist at the desired focal plane in tissue and a second waist at the fiber "output" of the mirror image probe. The merit function equally weighted both constraints to ensure a balance between spot size in tissue and back-coupling efficiency. After optimization the spot size (diameter) in tissue was estimated to be 13 tm in X and 14.5 plm in Y. The spot diagram at the focal plane in tissue is shown in Figure 6.5(B). Minimal astigmatism is observed due to the relatively low NA of the beam and relatively large curvature of the sheath. The estimated back-coupling into the fiber was estimated to be 47% after accounting for all losses and phase front curvature of the beam. 6.3.2 Minimization ofBackreflection For 3D-OCT imaging applications it is critical to minimize backreflected light from each surface in the probe. Backreflected light creates significant amounts of image noise by flooding the photodetector with incoherent photons. Backreflected intensities should generally be several orders of magnitude less than the expected reference arm power to minimize this effect. With typical power levels of -20 mW transmitted into the probe, backreflection levels of lower than -50 dB are required. Standard antireflection coatings provide -10 to -20 dB of reflection suppression and are not suitable for this application. Backreflection levels were reduced to -50 to -65 dB in the micro-optic lens system by angle polishing the end face of the glass ferrule and the first surface of the GRIN lens, and by using an epoxy spacer instead of an air spacer. An angle of +80 was introduced to the glass ferrule to reduce backreflection between the fiber/epoxy interface. The epoxy used in the spacer was chosen to have a refractive index of 1.524, which is very close to the refractive index of the fiber core (1.467). The first surface of the GRIN lens was polished at an angle of -80 in order to direct reflected light away from the fiber in the centre of the ferrule. The dominant source of remaining backreflection was the second surface of the angle prism, which could not be easily polished, since the beam exits the prism at close to normal incidence. In the future a custom angle prism with a slightly tilted output surface could be used to further minimize backreflection if necessary. 6.3.3 MeasuredPerformance The micro-optic lens design provided good performance and high assembly yields of better than 75%. Measurements of transmission, back coupling, working distance and backreflection levels are shown in Table 6.2. Working distance was measured in air from the output surface of the angle prism, and so does 120 not correspond to the desired working distance through water and tissue. A working distance of 940 Lm in air is desired for an optimal final working distance. Transmission levels were high (75%) and showed a very low standard deviation across the ten probes measured (+/- 3%). Back coupling was slightly higher than predicted by the ZEMAX model (59% vs. 47%), possibly due to the measurement being performed in air instead of water and tissue. The back coupling deviation was +/- 10%, which is slightly high but understandable given the sensitivity of back coupling to wavefront aberration. The working distance was 941 pm +/- 60.6 pm compared to an ideal value of 940 ptm. The working distance was precisely set for each individual probe by placing a mirror 940 pm from the lens tip and adjusting the axial position of the glass ferrule relative to the GRIN lens to maximize the back-coupled power. Backreflection levels were very low (-54 dB +/- 4.3 dB). Only one probe had to be discarded due to high backreflection (-44 dB). One additional probe was broken during connectorization, giving a total assembly yield of 80%. The combination of good performance and high yield made the bulk micro-optic lens system much more attractive than the fused fiber lenses for GI applications. Probe Number Transmission 1 2 3 4 5 6 7 8 9 10 MEAN STD 76% 81% 74% 77% 76% 74% 70% 74% 73% 78% 75% 3.0% Back Coupling 70% 74% 54% 66% 55% 64% 65% 44% 46% 55% 59% 10.0% Working Backreflection Distance [um] [dB] 914 -51 940 -57 -54 965 902 -54 -53 940 890 -44 -59 953 1041 -56 -58 840 1026 -52 -54 941 60.6 4.3 Table 6.2. Measured performance of ten micro-optic lens probes. One probe was rejected due to high backreflection levels (-44 dB) and one was broken during connectorization. 80% of the probes were useable. 6.4 Mechanical Design The mechanical design of the probe is equally important to the optical design, since stable transmission of the rotational and pullback motion from the proximal PIU to the distal tip is required to generate a high-quality spiral scan pattern of the imaging beam. In addition the probe must be designed to be flexible, robust, and small enough to fit down the working channel of a conventional endoscope (2.8 mm diameter). The following sections describe the mechanical design of the torque coil, proximal joint 121 between the torque coil and rigid hypo tube, and fiber connector / water flush port for the high-speed imaging probe. 6.4.1 Torque Coil Selection A flexible torque coil comprised of multiple layers of counter-wound wire is used to transmit rotational and pullback motion from the PIU to the distal tip of the probe. Torque coil designs inherently trade off stiffness against torque transmission; a very stiff coil accurately transmits rotation and translation, whereas a flexible coil is more compressible and therefore transmits motion less accurately. Two torque coil designs with different stiffness characteristics were tested for GI endoscopy applications. The more rigid design was found to translate push/pull and rotary motion very effectively, but the lack of flexibility caused binding when the endoscope was actuated. The more flexible design had much less severe binding problems and still produced highly stable rotational motion at up to 60 Hz. Push/pull translation was degraded compared to the rigid torque coil but overall performance was significantly better. The final torque coil used in the probe was a 3-layer design with an outer diameter of 1.45 mm and an inner diameter of 0.43 mm. The outer diameter of SMF-28 fiber with a standard coating is 0.25 mm so the fiber could be readily passed down the torque coil. The overall length of the torque coil was 2 m or 3 m, making the probes compatible with conventional GI endoscopes. 6.4.2 ProximalJoint A potential failure point is present at every location where two probe components are joined together. It is therefore important to design each joint for high strength while still allowing for high-speed translational and rotational motion. As described in Section 6.3, the joint between the micro-optic lens and torque coil is covered in a rigid hypo tube to provide strain relief and prevent separation of the lens from the coil. A similar method is employed at the proximal end of the probe. Here the torque coil joins a long, rigid hypo tube that is epoxied to a standard FC/APC fiber connector on the other end. The connector and hypo tube allow the probe to be attached to the PIU. The joint between the hypo tube and torque coil is shown in Figure 6.6. A protective hypo tube is epoxied over top of the torque coil and outer hypo tube that is attached to the FC/APC connector. A freely floating inner hypo tube is positioned over top of the SMF-28 fiber to prevent epoxy from adhering to the fiber and causing stress when the probe is flexed. A layer of heat shrink is attached to the outer hypo tube to improve centration within the protective hypo tube. This joint structure provides strain relief and high strength while still permitting the fiber to flex within the torque coil. 122 6.4.3 Fiber Connector and Flush Port An FC/APC connector and water flush port are attached to the proximal probe section as shown in Figure 6.7. A thin piece of heat shrink is placed on the outer hypo tube to ensure centration inside the FC/APC connector. Centration is critical to prevent off-axis motion of the probe when the PIU motor is rotating. A plastic adapter to secure the probe in the PIU is fastened over a metal sleeve, and a standard Tuohy-Borst flush port is used to allow water flushing of the probe during imaging. The plastic sheath containing the torque coil and fiber is joined to the flush port to form a tight seal. 6.4.4 Future ProbeDesign While initial imaging studies using the micro-optic probe were encouraging, further optimization of the design is possible. The overall outer diameter of the probe was 2.34 mm - 2.54 mm depending on the type of plastic sheath used, which is only slightly smaller than the nominal endoscope working channel inner diameter of 2.8 mm. Some probe binding and unstable push/pull actuation was therefore observed during some procedures when a tight radius bend was introduced into the endoscope. This binding was likely caused by deformation of the working channel cross-section into an ovular shape, which put pressure on the probe sheath and prevented uniform rotation and pullback. This problem will be addressed in a revised version of the probe by reducing the overall outer diameter to 1.80 mm (30% - 40% reduction). The revised probe is shown in Figure 6.8. Further miniaturization of the probe will be possible by using a thinner torque coil with an outer diameter of 1.00 mm, allowing the hypo tube holding the lens components to be extended back and used for strain relief. This allows removal of the second hypo tube previously used for strain relief between the torque coil and lens, and significantly reduces the overall diameter. A smaller angle prism is required to ensure that it will fit into the hypo tube. Additionally, the entire lens assembly can be sealed with IR transmissive epoxy to reduce the buildup of debris on the angle prism surface. The small change in the torque coil thickness is not expected to significantly impact the rotational stability of the probe, and push/pull stability should be improved by reducing pressure from the endoscope. 6.5 Imaging Performance Comparison The bulk micro-optic probe was tested in several normal human subjects during screening colonoscopy and the resulting 3D-OCT data was compared to earlier images acquired with a modified commercial fused fiber probe originally designed for cardiovascular applications. The micro-optic probe was found to provide significantly reduced rotational jitter due to the more robust torque coil. The pullback uniformity of the two probes was similar. The micro-optic probe also produced a higher quality 123 beam, tighter focus, higher back-coupling efficiency than the cardiovascular probe, resulting in improved signal to noise ratios and reduced transverse speckle noise. These features translated into significantly improved enface and cross-sectional 3D-OCT image quality in human subjects. A side-by-side comparison of enface image quality for data acquired using the bulk micro-optic probe and modified cardiovascular probe is shown in Figure 6.9. Both images were acquired in the sigmoid colon of normal patients and show transverse features consistent with colonic crypts. The significantly improved rotational stability of the micro-optic probe results in enhanced visualization of crypts. Both probes show some non-uniformity in pullback motion that cause crypts to appear elongated. A side-by-side comparison of radial image quality for data acquired using the bulk micro-optic and modified cardiovascular probe is shown in Figure 6.10. Both images were acquired in the human finger pad, although the image in Figure 6.10(A) also shows an infrared viewing card in the bottom-left quadrant. Both images were acquired at the same axial line rate and probe rotational speed, ensuring that the number of lines per image is the same for both cases. The better optical performance of the microoptic probe results in significantly improved transverse resolution, higher signal to noise ratio, and reduced speckle size compared to the cardiovascular probe. Overall, the micro-optic probe provides excellent optical performance and can be reliably used for GI endoscopy procedures. 124 6.6 Figures 1540 um 145 um 230 um I - '1 ,, 30 LAYOUT FUSED FIBER LENS WITH COMMERCIAL MULTIMODE SECTION THU TAN 15 2009 in IMagluI1*Auruml CONFIGURATION 50 lu- 1 OF um 1 Figure 6.1. ZEMAX model of a fused fiber lens system using commercial multimode fiber as the focusing element. A: Schematic diagram showing optimized layout. Order of elements is coreless fiber, multimode fiber, coreless fiber, and water. B: Spot diagram at focal plane. Estimated beam radius is 15 um at a working distance of 1540 um. 573 um 559 um FUSED FIBER LENS WITH THU TRN 15 2009 1540 um 200 um 3D LRYOUT CUSTOM MULTIMODE SECTION 40 um Figure 6.2. ZEMAX model of a fused fiber lens system using custom multimode fiber as the focusing element. A: Schematic diagram showing optimized layout. Order of elements is coreless fiber, multimode fiber, coreless fiber, and water. B: Spot diagram at focal plane. Estimated beam radius is 8 um at a working distance of 1540 um. CONFIGURATION 1 OF 1 To Proximal End Figure 6.3. Schematic diagram of fused fiber lens showing polish orientations on distal tip. An air gap formed by an ultrathin sheath and epoxy bead creates total internal reflection, directing the beam out of the lens at an 800 angle. C, coreless fiber. MM, multimode fiber. E, epoxy bead. 125 02.34 01.93 01.83 -01.27 GRIN Lens Epoxy IR Transmissive Epoxy Glass Ferrule Angle Prism Torque Coil Glass Fele Hypo Tube Plastic Sheath 18.00 Hypo Tube SMF-28 Fiber Water Desired Focal Plane . . .. Figure 6.4. Schematic diagram of bulk micro-optic lens system and distal tip of complete probe. All dimensions are in mm. Focal plane is set at 500 um from the sheath surface. Inset shows an end view of the probe. SEpoxy GRIN Lens Tissue Prism Water Sheath 43D Mirror Image - LAYOUT BULK MICRO-OPTIC LENS SYSTEM RNO PROXIMRL PROBE TIP THU TRN 15 2009 30 um CONFIGURATION 1 OF 1 Figure 6.5. ZEMAX model of a bulk micro-optic lens system. A: Schematic diagram showing optimized layout. Model is constructed as a mirror image to allow estimation of fiber back-coupling efficency. B: Spot diagram at focal plane. Estimated beam radius is 13 pm x 14.5 pm (XY) at a working distance of 500 pm in tissue. Epoxy Heat Shrink - Outer Hypo Tube Inner Hypo Tube Protective Hypo Tube SMF-28 Fiber Torque Coil Figure 6.6. Schematic diagram of the proximal joint between the torque coil and rigid hypo tube for the bulk micro-optic probe. A freely floating inner hypo tube is used to prevent epoxy from stressing the fiber. 126 Figure 6.7. Schematic diagram of the FC/APC connector and water flush port for the bulk micro-optic probe. 0 01.80 01.50 -01.27 01.07 Angle Prism 01.00 Torque Coil Epoxy IR Transmissive Epoxy Glass Ferrule Plastic Sheath 17.00 Hypo Tube GRIN Lens Water .. ... km.. Desired Focal Plane hypo tube is Figure 6.8. Schematic diagram of the future reduced-diameter micro-optic probe. A single torque coil. used to hold the lens elements and provide strain relief for the joint between the lens and the probe. the of view Overall outer diameter can be reduced to 1.80 mm. Inset shows an end face 127 Figure 6.9. Comparison of en face image quality for normal human colon acquired with a bulk micro-optic probe a modified cardiovascular probe. A: En face image obtained using a bulk micro-optic probe. B: En face image obtained using a modified cardiovascular probe. The micro-optic probe provides improved rotational stability, resulting in improved visualization of transverse tissue features such as colonic crypts. Nonuniform pullback motion in both probes causes occasional image artifacts (arrows). Figure 6.10. Comparison of radial image quality for human finger pad acquired with a bulk micro-optic probe a modified cardiovascular probe. A: Image obtained using a bulk micro-optic probe. B: Image obtained using a modified cardiovascular probe. The micro-optic probe provides improved beam quality, spot size, and back-coupling efficiency, resulting in improved contrast and resolution. 128 6.7 References [1] W. A. Reed, M. F. Yan, and M. J. Schnitzer, "Gradient-index fiber-optic microprobes for minimally invasive in vivo low-coherence interferometry," Optics Letters, vol. 27, pp. 17941796, Oct 15 2002. [2] M. S. Jafri, S. Farhang, R. S. Tang, N. Desai, P. S. Fishman, R. G. Rohwer, C. M. Tang, and J. M. Schmitt, "Optical coherence tomography in the diagnosis and treatment of neurological disorders," Journalof Biomedical Optics, vol. 10, pp. -, Sep-Oct 2005. [3] H. Li, B. A. Standish, A. Mariampillai, N. R. Munce, Y. X. Mao, S. Chiu, N. E. Alarcon, B. C. Wilson, A. Vitkin, and V. X. D. Yang, "Feasibility of interstitial Doppler optical coherence tomography for in vivo detection of microvascular changes during photodynamic therapy," Lasers in Surgery and Medicine, vol. 38, pp. 754-761, Sep 2006. [4] Y. X. Mao, S. Chang, S. Sherif, and C. Flueraru, "Graded-index fiber lens proposed for ultrasmall probes used in biomedical imaging," Applied Optics, vol. 46, pp. 5887-5894, Aug 10 2007. 129 CHAPTER 7 7.0 Clinical 3D-OCT in the Upper Gastrointestinal Tract 130 7.1 Motivation Endoscopic therapies are becoming increasingly common in gastroenterology due to their less invasive nature, lower morbidity, and faster recovery time compared to most surgical interventions. Endoscopic therapies, however, are not indicated for every patient and multiple treatments may be required to obtain adequate control of a disease. Gastroenterologists can therefore benefit from imaging tools that can aid in pre-interventional planning and follow-up assessment of endoscopic therapies. Follow-up assessment can include both determining the completeness of treatment and checking for recurrence of disease. 3D-OCT is desirable for these applications since it is compatible with existing commercial endoscopes and enables comprehensive depth-resolved tissue imaging. Advanced image processing techniques enable improved visualization of tissue microstructure for assessing features beneath the superficial layers typically apparent under endoscopy. These techniques include the generation of cross-sectional images with arbitrary orientations, generation of projection views similar to en face microscopy images, improved quantitative measurements of morphology, 3D image processing methods for speckle noise reduction, and virtual manipulation of tissue geometry for visualizing structural morphology. Barrett's esophagus (BE) is a pathology of the upper gastrointestinal (GI) tract that is associated with an approximately 40-fold increases in risk of progression to dysplasia and adenocarcinoma compared to the general population. Recently BE has become amenable to new endoscopic therapies to treat the disease prior to progression to invasive cancer. BE results from chronic mucosal injury and is a precursor condition to esophageal adenocarcinoma [1]. Esophageal cancer has a five-year survival rate of only 16% [2], but early detection and treatment achieves a high percentage of regression in patients with dysplasia [3]. BE is characterized by the replacement of squamous epithelium with specialized intestinal columnar epithelium [4]. Neoplastic changes in BE develop in stages from non-dysplastic metaplasia to increasing grades of dysplasia and eventually to adenocarcinoma [5]. Differentiation of normal mucosa, BE, dysplasia, and carcinoma are necessary for an imaging method to be used to assist pre- or post-therapy assessment. Despite initial enthusiasm over chromoendoscopy (using chemical dyes to increase contrast), narrow band imaging, autofluorescence, and other imageenhancing methods, larger studies have shown poor discrimination between BE and early-stage cancers. A technology such as 3D-OCT that can be used for guiding excisional biopsy, providing subsurface tissue imaging, planning endoscopic therapies, and assessing patients at follow-up would significantly improve overall outcomes. Currently, random four-quadrant biopsies taken every 1-2 cm along the region of BE (Seattle protocol) is the clinical standard for detecting dysplasia and adenocarcinoma during BE surveillance [6, 7]. This procedure suffers from high false negative rates due to sampling errors, since high-grade dysplasia and early invasive cancer are typically not visible or distinguishable from 131 surrounding BE by standard endoscopy. Moreover, it is known from surgically resected specimens that both dysplastic and cancerous lesions can be multiple, scattered, and small, thereby making detection of these lesions especially prone to sampling errors. The prevalence of undetected adenocarcinoma from the random biopsy protocol ranges from 33-57% [8-11]. Radiofrequency ablation (RFA) has recently been introduced to treat diffuse conditions such as BE [12]. RFA produces a broad, superficial ablation pattern compared to the deep, focal ablation obtained with argon plasma coagulation, electrocautery, or heater probes. Endoscopic 2D-OCT analysis has shown that some patients can experience recurrence of BE following RFA treatment [13] [14, 15], which could be associated with a future risk of adenocarcinoma under the neosquamous epithelium [16, 17]. Endoscopic mucosal resection (EMR) is a common method for treating visible dysplastic nodules arising in a setting of BE. In EMR, fluid is typically injected into the submucosa to raise the nodule. The nodule is then removed with an electrocautery snare with or without the aid of a resection cap or lift-banding. EMR is also associated with some disease recurrence due to incomplete extraction of dysplastic lesions. This motivates the need for 3D-OCT endomicroscopy over larger fields of view with complete visualization of 3D tissue microstructure in order to assess endoscopic therapies in the planning stage and at follow-up. In this section of the thesis work, a 3D-OCT endomicroscopy system and spiral-scanning imaging catheter were used to assess five tissue categories: 1) no pathology ("Normal"); 2) BE prior to undergoing radiofrequency ablation therapy ("BE pre-RFA"); 3) sites previously treated with radiofrequency ablation therapy ("BE post-RFA"); 4) visible esophageal nodules prior to undergoing endoscopic mucosal resection ("Nodule pre-EMR"); and 5) sites previously treated with endoscopic mucosal resection ("Nodule post-EMR"). 3D-OCT was used to establish characteristic features of esophageal mucosa for each tissue group. 3D-OCT was found to be a valuable adjunct for endoscopic therapies such as radiofrequency ablation (RFA) and endoscopic mucosal resection (EMR) for treatment of BE and esophageal nodules, enabling tissue characterization over surface areas many times larger than the typical size of excisional biopsy specimens. 7.2 Clinical Protocol All subjects in this study were imaged at the Veterans Affairs Boston Healthcare System (VABHS) Jamaica Plain campus in collaboration with Dr. Hiroshi Mashimo, MD PhD, and Dr. Qin Huang, MD. Subjects were recruited from the pool of patients undergoing elective endoscopy for screening, surveillance, or treatment of BE or nodules. Subjects followed a standard preparation procedure for upper endoscopy, including cessation of blood thinners and fasting prior to the appointment. 3D-OCT imaging was performed in tandem with standard video endoscopy. 132 For normal subjects, the endoscopist identified regions for 3D-OCT imaging that appeared normal and unremarkable using white light video endoscopy. Imaging was performed by inserting the 3D-OCT catheter down the working channel of a standard endoscope. For pre-treatment subjects, the endoscopist identified locations suspicious for BE or esophageal nodules using changes in tissue surface texture and colour visible under white light video endoscopy and narrow band imaging (NBI). For post-treatment subjects, the endoscopist identified regions of prior treatment using patient history notes and, in some cases, subtle visual cues that differentiate treated tissue from untreated tissue. Where clinically indicated, pinch biopsy samples were obtained from the 3D-OCT imaging sites following the conclusion of all image acquisition. Between 1-6 sites were assessed for each subject, increasing the total procedure time by an average of 10.5 minutes and by no longer than 19 minutes. This increase in procedure length is similar to the time increase associated with chromoendoscopy [18]. After the acquisition of all 3D-OCT datasets, the imaging catheter was withdrawn from the working channel and replaced with pinch biopsy forceps to acquire biopsies of the imaged tissue where clinically indicated. Biopsies were generally not obtained from normal subjects due to the slight risk associated with pinch biopsy and the lack of clinical necessity. Histopathology from pinch biopsies was evaluated by an expert blinded pathologist in order to establish a tissue classification for each site. The histological diagnoses were then compared to tissue features based on the 3D-OCT data. In all cases, the 3D-OCT endomicroscopy system described in Chapters 5 - 7 was used to obtain volumetric image data. While imaging parameters varied slightly due to system upgrades (such as FDML sweep linearization) and continuing probe development, typical performance is as described in Chapter 5. To briefly reiterate, sensitivity was 105 - 107 dB with 13 - 15 mW of power incident on the tissue. Optical resolution in tissue was 4.7 - 5.1 Rim (axial) and 10 - 15 [tm (transverse). Maximum imaging ranges were 1600 - 1800 gpm from the centre of the probe, although the useable attenuation-limited imaging depth was typically 1000 - 1200 [tm from the tissue surface. True 3D resolution varied with probe rotation speed and pullback speed, but the system was typically set to provide axial line spacing of 7 - 12 [tm and frame-to-frame spacing of 8 - 12 pm. Axial pixel spacing was typically 3.0 - 3.7 tm, and varied depending on the total imaging range. In total, 61 sites from 18 unique patients were imaged during this study from February to December 2008. Table 7.1 shows a summary of the collected 3D-OCT image data. Note that the total number of patients in Table 7.1 exceeds 18 since multiple tissue types were imaged from several patients, typically consisting of a normal region and a pathological region. The "other" grouping includes pathologic tissue that was outside the main scope of the study, including ulcers, strictures, gastric anastomosis, and 133 eosinophilic esophagitis. The following sections show representative 3D-OCT data, characteristic features, and comparisons to histology for each tissue type. Tissue Type Normal Esophagus BE Pre-RFA BE Post-RFA Nodule Pre-EMR Nodule Post-EMR Other Pathology Number of Patients 7 5 7 3 2 7 Number of Imaging Sites 11 11 12 8 5 14 Table 7.1. Summary of 3D-OCT data collected during upper endoscopy. 7.3 Normal Esophagus Obtaining normal image data was a necessary first step in this study in order to establish a baseline for comparison to pathologic tissue. 3D-OCT images of normal esophageal mucosa were obtained at 11 sites from 7 unique patients. All patients tolerated the procedure well and there were no known immediate or long-term complications. Useable data was obtained for 5 out of the 7 patients. 2 patients produced unusable data due to severe motion artifacts from breathing, heartbeat, gagging, or whole-body motion during 3D-OCT imaging. Results of the normal esophagus imaging are described in Sections 7.3.1 to 7.3.2 below. 7.3.1 CharacteristicFeatures Figure 7.1 shows typical 3D-OCT data obtained from the normal esophagus. Figure 7.1(A) shows a volumetric rendering of the entire 980 x 512 x 1268 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 7.5 mm, and the imaging range was 1.7 mm in tissue. The images in Figure 7.1 have been axially cropped to remove the probe sheath and non-useable points beyond the penetration depth of light in the tissue. The useable imaging depth in tissue was approximately 1100 jim. Unless otherwise specified, all crosssectional and en face images shown below are formed by averaging the volumetric data over a 20 !imthick section to reduce speckle noise. Since 20 lim is approximately the size of a single epithelial cell, minimal image blurring is observed from this process [19]. The red, green, and blue boxes in Figure 7.1(A) illustrate the locations of orthogonal cut planes used to extract en face and cross-sectional image slices for closer examination. In this and all subsequent 3D-OCT image data the X axis represents the probe's rotational axis (fast, 50 - 70 Hz), the Y axis represents the axial OCT line data, and the Z axis represents the probe's pullback direction (slow, 500 - 1000 jim/s). 134 Figure 7.1(B) shows an en face image centered at a tissue depth of 175 tm, corresponding to the centre of the squamous epithelial layer. This image is similar in concept to a conventional microscopy image obtained by setting the focus to a depth of 175 jtm. Consistent with previous 2D-OCT studies [20] the normal squamous epithelium is featureless and unremarkable. The small, scattered hypointense regions in this image are artifacts caused by debris on the probe surface. Slight transverse jitter oriented primarily along the X axis is due to patient heartbeat and respiration artifacts. The blue and green arrow pairs on the border of Figure 7.1(B) indicate the positions of the cross-sectional images shown in Figure 7. 1(C,D). One of the main benefits of 3D-OCT endomicroscopy compared to 2D-OCT is the ability to align cross-sectional images to enface tissue features identified in XZ images. Figure 7.1(C) shows a YZ crosssectional image at the position indicated by the green arrow pair in Figure 7. I1(B). A regular, well-defined layered architecture can be appreciated throughout the length of the pullback. A slight discontinuity is present at -14 mm due to sudden patient motion. The red arrow pair indicates the position of the en face image in Figure 7.1(B). Figure 7.1(D) shows an XY cross-sectional image at the position indicated by the blue arrow pair in Figure 7.1(B). The regular layered architecture is visible in XY as well. In both crosssections, the epithelium is devoid of glands or vessels. The lamina propria and submucosa tissue layers (first and second hyperintense layers, see Figure 7.3) contain small (< 100 tm) regular blood vessels, identifiable by their axial shadows and interconnectivity across numerous cross-sectional images. The submucosa also contains scattered, non-shadowing glandular structures that may represent normal seromucous glands. The ability to definitively distinguish vascular from glandular features without the use of Doppler processing is unique to 3D-OCT. This ability is valuable in the assessment of pathologic tissue, as discussed below. The dense spatial sampling density, high optical resolution, and large field of view of 3D-OCT endomicroscopy enables high-magnification analysis of each tissue layer over arbitrary regions of interest (ROI). This principle is demonstrated in Figure 7.2. Figure 7.2(A-D) shows en face images at the ROI shown as a black dashed box in Figure 7.1(B). Images are shown at tissue depths of 175, 340, 420, and 520 p~m for Figure 7.2(A-D) respectively. These depths correspond to the epithelium, lamina propria, muscularis mucosa, and submucosa respectively. Regular vascular networks are visible in both the lamina propria and submucosal layers. Individual vessels can be identified in the en face images (red and black triangles in Figure 7.2(B,D)) as well as the cross-sectional images shown in Figure 7.2(E-G). Nonvascular glands are small (< 100 tm), regular, and confined to the muscularis mucosa and submucosa. 135 7.3.2 Comparison to Histology Comparison of 3D-OCT endomicroscopy images to histology slides obtained from excisional biopsy and to white light video endoscopy examination is a good method for linking image features to known tissue structures. Figure 7.3(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 7.3.1. The esophageal mucosa appears pale-pink, smooth, and regular, consistent with normal squamous tissue. Figure 7.3(B) shows representative histology of normal esophageal squamous mucosa. Since there was no clinical indication to biopsy this particular patient, the histology image in Figure 7.3(B) is taken from another subject where biopsy was required in the midesophagus but a normal diagnosis was made by the pathologist. Excellent correlation is shown between the histology image and the 3D-OCT image. The epithelium, lamina propria, muscularis mucosa, submucosa, and muscularis propria are visible as well-separated and distinct layers with alternating hypoand hyperintensity under 3D-OCT (Figure 8.3(C)). The pinch biopsy sample did not extend deeper than the lamina propria, highlighting another advantage of 3D-OCT compared to conventional histology. Small, regular vessels that produce axial shadows are visible (V) as well as non-shadowing submucosal glands (G) under 3D-OCT. In total, 5 out of the 5 normal patients (100%) with useable 3D-OCT data showed features consistent with those described above. A normal diagnosis was histologically confirmed in 2 out of 2 patients (100%) where biopsy specimens were available. The remaining 3 patients did not have biopsy samples available due to lack of clinical indication. Moderate variability was observed in the apparent tissue layer thicknesses from patient to patient, and also at multiple sites within the same patient. In general the variability was uniform across each of the mucosal layers, suggesting that this is due to varying tissue compression from endoscope suction and changes in probe contact pressure with the luminal surface. Similar issues have also been observed during 2D-OCT endoscopy and confocal endoscopic microscopy. 7.4 Barrett'sEsophagus: Pre-Radiofrequency Ablation 3D-OCT images of BE were obtained at 11 sites from 5 different patients. This data was collected in order to establish the characteristics of active BE and allows a comparison between untreated and RFAtreated BE. 3 of these patients had not previously been treated with RFA, EMR, or other endoscopic or surgical therapies for BE. 1 patient was previously treated with EMR 3 months prior to 3D-OCT imaging to remove a dysplastic nodule present within a BE region. 1 patient was treated with RFA 4 months prior to 3D-OCT imaging to ablate a region of BE, but presented on the imaging day with residual BE. All 5 patients had been medically treated with proton pump inhibitors or other pharmaceuticals to control BE progression. All patients tolerated the procedure well and there were no known immediate or long-term 136 complications. Useable data was obtained for all patients. Results of the BE pre-RFA imaging are described in Sections 7.4.1 to 7.4.2 below. 7.4.1 CharacteristicFeatures Figure 7.4 shows typical 3D-OCT data obtained from a BE region prior to any endoscopic or surgical therapy. This patient presented with small BE islands in the distal esophagus near the gastro-esophageal junction (GE junction). BE islands represent a clinical situation where 3D-OCT imaging could be particularly beneficial, since small isolated pockets of BE surrounded by normal squamous mucosa may be more difficult to discern than the more typical continuous BE "fingers" extending proximally from the GE junction. BE may also present as short segment, ultra-short segment, or invisible lesions at the GE junction. These cases may also represent situations where 3D-OCT can provide enhanced visualization over endoscopic examination alone. Figure 7.4(A) shows a volumetric rendering of the entire 980 x 512 x 1268 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 7.5 mm, and the imaging range was 1.7 mm in tissue. The red, orange, green, blue, and purple boxes in Figure 7.4(A) illustrate the locations of orthogonal cut planes used to extract enface and cross-sectional image slices. Figure 7.4(B) shows an en face image centered at a tissue depth of 175 pm, corresponding to the centre of the squamous epithelial layer. The probe was pulled back over the GE junction during acquisition of this dataset and a clear delineation can be appreciated between gastric mucosa on the lefthand side of the image (distal) and esophageal mucosa on the right (proximal). The gastric mucosa is covered with a regular pit pattern whereas the esophageal tissue appears relatively featureless. Figure 7.4(C) shows a second en face image at a tissue depth of 345 jpm. Several discrete regions of the esophageal tissue reveal striking atypical glandular structures consistent with BE. Glands are ovular or round in cross-section and range in length from 200 - 600 [pm. The BE islands are surrounded by comparatively normal-appearing squamous mucosa, illustrating the difficulty in analyzing these regions with 2D-OCT or random excisional biopsy. The blue, green, and purple arrow pairs on the border of Figure 7.4(B,C) indicate the positions of the cross-sectional images shown in Figure 7.4(D-F). Figure 7.4(D-F) shows cross-sectional images at multiple positions within the dataset. In Figure 7.4(D) normal gastric mucosa can be identified by regular vertical pit patterns and low penetration depths, consistent with previous 2D-OCT studies [20]. A -3 mm length of normal squamous mucosa with regular layered architecture is adjacent to a -5 mm length of BE tissue. In cross-section, the BE island exhibits distortion of the normal layered architecture and the presence of discrete hypointense glands. "Distortion" refers to localized, atypical variations in mucosal layer thicknesses as well as changes in layer-to-layer contrast compared with normal esophageal tissue. An orthogonal cross-section through a region 137 containing transitionary gastric cardia tissue and normal gastric mucosa is shown in Figure 7.4(E). This area appears to be free of BE. In comparison, a cross-section at a more proximal location as shown in Figure 7.4(F) is clearly afflicted by BE. These images illustrate that the BE glands are located beneath 300 - 500 pm of overlying tissue. A large number of glands are present along with distorted layered architecture. Figure 7.5 shows an enlarged view of the ROI illustrated in Figure 7.4(B,C). This ROI spans a portion of one BE island as well as an adjacent normal area. There is clear demarcation between the normal region (left) and BE island (right). Cross-sectional images in the ROI highlight the difference in layered appearance between the normal and BE tissue regions. The presence of glands and layer distortion are hallmarks of BE. Previous 2D-OCT studies have not conclusively established criteria for differentiating BE with low-grade dysplasia (LGD) from BE without LGD, although the gland density and degree of layer disruption suggested a more severe form of BE based on the 3D-OCT data. The intact submucosa indicates that HGD or adenocarcinoma is not present. 7.4.2 Comparison to Histology Figure 7.6 shows a comparison of 3D-OCT data to conventional white light video endoscopy and histology analysis of the BE island region. Figure 7.6(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 7.4.1. The esophageal mucosa contains discrete regions that appear red, inflamed, and irregular, consistent with BE islands. Figure 7.6(B) shows histology of BE islands taken near the GE junction of the same patient following 3D-OCT imaging. Excellent correlation is shown between the histology image and the 3D-OCT image. Numerous large BE glands buried beneath a layer of squamous tissue are visible in both the histology and 3D-OCT images with distortion of the regular layered architecture. This patient was diagnosed with BE without dysplasia after histology analysis, compared to dysplasia with LGD from the 3D-OCT data. This discrepancy is not surprising given the variations in both BE gland density and layered architecture visible in Figure 7.4(C). Gland density varies significantly over the 3D-OCT image volume and could easily be classified as BE without dysplasia if a single, less glandular region were used for analysis. This highlights an advantage of 3D-OCT for characterizing BE. Unlike 2D-OCT or excisional biopsy, 3D-OCT endomicroscopy enables a large surface area of 155 mm 2 to be comprehensively assessed for pathology, potentially reducing diagnostic sampling errors. In total, 4 of the 5 BE pre-RFA patients (90%) with useable 3D-OCT data showed features consistent with those described above. 1 of the 5 patients showed distortion of regular layered architecture without the development of clear glandular structure. A significant degree of variability was observed in gland size, gland packing density, and mucosal layer distortion from patient to patient. This was also observed 138 in previous larger-scale studies of BE with 2D-OCT [20] and is indicative of the heterogeneity of the BE pathology. Using the 3D-OCT data, 2 patients showed evidence of BE with LGD (both diagnosed as BE without dysplasia under histology) and 2 showed evidence of BE without dysplasia. For both tissue sites classified as BE without dysplasia based on 3D-OCT data, no histology was available due to RFA treatment and increased risk of bleeding on the day of imaging. 7.5 Barrett's Esophagus: Post-Radiofrequency Ablation 3D-OCT images of regions previously containing BE that had been treated with RFA were obtained at 12 sites from 7 unique patients. All 7 patients had received RFA at least once and at most five times prior to 3D-OCT imaging. All patients tolerated the procedure well and there were no known immediate or long-term complications. Useable data was obtained for 6 out of the 7 patients. 1 patient produced unusable data due to severe motion artifacts from breathing, heartbeat, gagging, or whole-body motion during 3D-OCT imaging. Results of the BE post-RFA imaging are described in Sections 7.5.1 to 7.5.2 below. 7.5.1 CharacteristicFeatures Figure 7.7 shows typical 3D-OCT data obtained from a region previously afflicted with BE that had been treated with RFA. This particular patient received his last RFA treatment 6 months prior to 3D-OCT imaging. Under white light video endoscopy the esophageal mucosa appeared regular and unremarkable, indicating that the RFA treatment was successful and that the BE had resolved. Assessment of previously treated BE regions represent another clinical situation where 3D-OCT imaging could be beneficial, since small regions of residual or recurrent BE surrounded by normal squamous mucosa may be extremely challenging to detect. Figure 7.7(A) shows a volumetric rendering of the entire 736 x 512 x 1613 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 8.6 mm, and the imaging range was 1.9 mm in tissue. The red, orange, green, and blue boxes in Figure 7.7(A) illustrate the locations of orthogonal cut planes used to extract en face and cross-sectional image slices for closer examination. Figure 7.7(B) shows an en face image centered at a tissue depth of 105 tm, corresponding to the centre of the squamous epithelial layer. The probe was pulled back over the GE junction during acquisition of this dataset and a clear delineation can be appreciated between gastric mucosa on the lefthand side of the image and esophageal mucosa on the right. The distinct linear border of the GE junction as visualized in Figure 7.7(B) may be a result of the previous RFA treatment, which uses a rectangular electrode to produce superficial tissue ablation. The gastric mucosa is, as expected, covered with a regular pit pattern whereas the squamous esophageal tissue appears relatively featureless. Figure 7.7(C) shows a 139 second en face image at a tissue depth of 390 jim. While the tissue appears largely normal, there are a small number of scattered glandular structures consistent with BE as indicated in Figure 7.7(C). Some hypointense features that appear to be glandular from this single en face plane are actually regular submucosal vessels, which was verified by vertically scanning the en face plane and observing interconnection of the features. The BE glands are isolated and surrounded by normal squamous mucosa, illustrating the difficulty in analyzing these regions with 2D-OCT or random excisional biopsy. The blue and green arrow pairs on the border of Figure 7.7(B,C) indicate the positions of the cross-sectional images shown in Figure 7.7(D,E). Figure 7.7(D,E) shows cross-sectional images at multiple positions within the dataset. In Figure 7.7(D), normal gastric mucosa and relatively normal squamous mucosa can be readily distinguished. An orthogonal cross-section through a region of normal gastric mucosa and isolated BE glands is shown in Figure 7.7(E). In cross-section, the isolated BE glands appear to be buried beneath regular epithelial and lamina propria layers. These glands can be differentiated from normal seromucous glands by their larger size, ovular cross-section, and position beneath the lamina propria. Figure 7.8 shows an enlarged view of the ROI illustrated in Figure 7.7(B,C). This ROI covers a region containing what appears to be a residual BE gland. Cross-sectional images in the ROI suggest that normal epithelium (first hypointense layer) has regrown over top of the BE gland. The lamina propria (first hyperintense layer) appears slightly thicker than normal, and there is not good demarcation between the laminia propria and muscularis mucosa. The lamina propria also contains subtle horizontally-oriented hyperintense striations. These striations could be fibrotic filaments formed as a result of healing following RFA treatment. The BE gland is ovular and approximately 300 x 85 jim in cross section. Compared to the untreated BE case presented in Section 7.4, the RFA-treated subject has significantly fewer BE glands. The few scattered glands that do remain are buried beneath a thicker layer of normal-appearing tissue. Buried glands are known to occur in some fraction of patients who have undergone ablative therapy for BE, although the exact prevalence is not well understood and seems to vary significantly according to the treatment type. Buried glands are thought to have reduced malignant potential compared to BE glands exposed to the esophageal lumen, but the chance for progression to dysplasia and adenocarcinoma is likely not reduced to zero. Buried glands are difficult to detect with endoscopic examination and pinch biopsy due to their focal and subepithelial nature. 3D-OCT could be particularly well-suited for detection and characterization of buried glands due to its large field of view and depth-resolved imaging capabilities. 140 7.5.2 Comparison to Histology Figure 7.9 shows a comparison of 3D-OCT data to conventional white light video endoscopy and histology analysis of the region previously treated with RFA. Figure 7.9(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 7.5.1. The esophageal mucosa appears regular and is similar to the image of normal esophagus shown in Figure 7.3. Figure 7.9(B) shows histology of a region near the GE junction of the same patient following 3D-OCT imaging. A discrepancy is observed between the histology image and the 3D-OCT image. Discrete glands buried beneath a normal epithelium and lamina propria are visible in 3D-OCT images but not the histology image. The size, shape, and hypointensity of the glands visible under 3D-OCT are highly suggestive of buried BE glands beneath neo-squamous epithelium although these findings could not be confirmed by pinch biopsy. The biopsy specimen was captured very close to the gastric mucosa and appears to show gastric/squamous transitional tissue. This tissue region was classified as normal GE junction after analysis of the histology data. Again, given the sparseness of the BE glands observed in the 3D-OCT image, it is not surprising that a single biopsy specimen would randomly sample a normal area. In total, 2 out of the 6 BE post-RFA patients (50%) with useable 3D-OCT data showed features consistent with completely regular squamous epithelium and showed no signs of residual or recurrent BE in regions that appeared endoscopically normal (no histology available for both patients). 4 out of 6 patients (67%) showed some signs of residual BE in regions that appeared endoscopically normal as described above (2 BE under histology, 1 normal under histology, no histology available for 1 patient). The situation of non-visually apparent, scattered, buried BE should be differentiated from patients presenting for follow-up after RFA where large untreated regions of BE are visible endoscopically. This situation is relatively common and is a result of incomplete ablation with RFA, and is rectified by further RFA treatments to the BE site. The clinical implications of occult, sparse buried BE as implicated in Figure 7.7 - Figure 7.9 are not entirely clear. It is possible that, if left untreated, the buried BE areas could lead to recurrent wide-spread BE or even progress to dysplasia. The sparse gland distribution would make detection and treatment of residual BE extremely challenging with conventional endoscopic techniques, suggesting that 3D-OCT could play a role in follow-up assessment of RFA for BE. 7.6 Esophageal Nodules: Pre-Endoscopic Mucosal Resection 3D-OCT images of esophageal nodules in the presence of BE were obtained at 8 sites from 3 unique patients. 1 patient had not previously been treated with RFA, EMR, or other endoscopic or surgical therapies for BE or nodules. 1 patient was previously treated with RFA prior to 3D-OCT imaging to treat the underlying BE condition. 1 patient had previously undergone EMR at a different site in the esophagus to remove a previous dysplastic nodule. EMR was attempted on all 3 patients following 3D-OCT 141 imaging, with 2 out of 3 resections concluded successfully. I EMR attempt failed due to fibrosis at the nodule site. All patients tolerated the 3D-OCT imaging procedure well and there were no known immediate or long-term complications. Useable data was obtained for 2 out of the 3 patients. 1 patient produced unusable data due to kinking of the probe sheath and subsequent severe frame-to-frame image jitter. Results of the nodule pre-EMR imaging are described in Sections 7.6.1 to 7.6.2 below. 7.6.1 CharacteristicFeatures Figure 7.10 shows typical 3D-OCT data obtained from an esophageal nodule prior to EMR resection. This particular patient was not previously treated with RFA or EMR. Under white light video endoscopy the esophageal mucosa appeared red and irregular with the nodule clearly visible. EMR to remove esophageal nodules is not clinically indicated if the nodule is malignant and has invaded through the submucosa, so 3D-OCT imaging could be particularly beneficial for assessment prior to EMR. Figure 7.10(A) shows a volumetric rendering of the entire 736 x 512 x 1919 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 7.0 mm, and the imaging range was 1.6 mm in tissue. The red, orange, green, and blue boxes in Figure 7.10 (A) illustrate the locations of orthogonal cut planes used to extract enface and cross-sectional image slices. Figure 7.10 (B) shows an en face image centered at a tissue depth of 175 jim, corresponding to the centre of the squamous epithelial layer. The probe was pulled back over the GE junction during acquisition of this dataset and a clear delineation can be appreciated between gastric mucosa on the lefthand side of the image and esophageal mucosa on the right. The gastric mucosa is, as expected, covered with a regular pit pattern. The esophageal tissue at the far right appears relatively featureless. A -3 mm section of tissue between the gastric and normal squamous regions (indicated by the black dashed box in Figure 7.10(B)) contains highly irregular glandular structures. Figure 7.10(C) shows a second en face image at a tissue depth of 450 gm. The glands extend down at least 275 gm and branching is observed at deeper positions. The blue and green arrow pairs on the border of Figure 7.10(B,C) indicate the positions of the cross-sectional images shown in Figure 7.1 0(D,E). Figure 7.10(D,E) shows orthogonal cross-sectional images through the highly glandular region. In Figure 7.10(D) normal gastric mucosa and relatively normal squamous mucosa are flanking the glandular region. Based on the branching, irregular structure of the glands this region is assumed to be the esophageal nodule. An orthogonal cross-section as shown in Figure 7.10(E) confirms the highly irregular nature of the glands. Disruption of layered architecture is apparent. These glands can be differentiated from the more typical BE glands by their larger size, extension from the luminal surface to > 500 im tissue depths, and distinct branched appearance. These 3D-OCT features have, to the author's knowledge, 142 not been previously observed. Figure 7.11 shows an enlarged view of the ROI illustrated in Figure 7.10(B,C). This ROI covers a portion of the nodule region. Cross-sectional images in the ROI reveal loss of layered architecture and concentrated branching glands. No distinct submucosal layer is present, suggesting the possibility of invasion through to the muscularis propria consistent with malignant disease. Previous 2D-OCT studies have not reported this type of branching glandular structure in HGD or adenocarcinoma arising in a background of BE, however, so tissue classification based on the 3D-OCT data is challenging. 7.6.2 Comparison to Histology Figure 7.12 shows a comparison of 3D-OCT data to conventional white light video endoscopy and histology analysis of the esophageal nodule. Figure 7.12(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 7.6.1. The nodule is clearly visible under white light examination. Figure 7.12(B) shows histology of the nodule following excision with EMR. Good correlation is apparent between the histology image and the 3D-OCT image. Large, irregular glands are visible in both the histology and 3D-OCT image with clear and widespread loss of the normal squamous layered architecture. This patient was diagnosed with BE with squamoid metaplasia of the mucin glands after analysis of the histology image. In total, 1 out of the 2 nodule pre-EMR patients (100%) showed evidence of HGD under 3D-OCT imaging (confirmed with histology). 1 patient, as described above, could not be classified using 3D-OCT due to lack of comparable data (squamoid metaplasia under histology). While a larger study is required to accurately determine the ability of 3D-OCT to stage esophageal nodules prior to EMR, these initial results are encouraging. 7.7 Esophageal Nodules: Post-Endoscopic Mucosal Resection 3D-OCT images of regions previously treated with EMR to remove esophageal nodules were obtained at 5 sites from 2 unique patients. All patients tolerated the 3D-OCT imaging procedure well and there were no known immediate or long-term complications. Useable data was obtained for 2 out of the 2 patients. Results of the nodule post-EMR imaging are described in Sections 7.7.1 to 7.7.2 below. 7. 7.1 CharacteristicFeatures Figure 7.13 shows typical 3D-OCT data obtained from a region previously treated with EMR to remove an esophageal nodule. This particular patient had EMR to remove a nodule with HGD approximately 2.5 months prior to 3D-OCT imaging. Under white light video endoscopy the treated region appeared relatively normal except for a raised tissue ridge on the border of the EMR site. Previous 143 biopsy results from the EMR borders were positive for dysplasia although this could not be visually confirmed under endoscopic examination. 3D-OCT analysis could be a useful tool to search the prior EMR site for suspicious tissue, improving biopsy guidance and reducing sampling errors. Figure 7.13(A) shows a volumetric rendering of the entire 980 x 512 x 1268 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 7.6 mm, and the imaging range was 1.7 mm in tissue. The red, orange, green, and blue boxes in Figure 7.13(A) illustrate the locations of orthogonal cut planes used to extract enface and cross-sectional image slices. Figure 7.13(B) shows an en face image centered at a tissue depth of 175 [tm, corresponding to the centre of the squamous epithelial layer. The probe was pulled back over the GE junction during acquisition of this dataset. The GE junction is less apparent here than in previous en face images due to poor probe contact with the gastric mucosa. The esophageal epithelium appears relatively featureless and regular at this tissue depth. The three dark bands running parallel to the pullback axis are polarization artifacts caused by probe rotation and are not clinically relevant. Figure 7.13(C) shows a second enface image at a tissue depth of 350 jtm. The border of the EMR site can be distinguished by a sharp change from hyperintense tissue (possibly fibrosis) inside the EMR site to more hypointense tissue outside. A small section of tissue (indicated by the dashed box in Figure 7.13(B,C)) at the distal EMR margin contains small glandular structures consistent with BE, but no large irregularities consistent with invasive disease or HGD are visible. The blue and green arrow pairs on the border of Figure 7.13(B,C) indicate the positions of the cross-sectional images shown in Figure 7.10(D,E). Figure 7.13(D,E) shows orthogonal cross-sectional images through the glandular region. In Figure 7.13(D), normal gastric mucosa and normal squamous mucosa are flanking the prior EMR site scar and the region suspicious for BE, but no features indicative of HGD or invasive disease are visible. The scar is characterized by hyperintense tissue layers of possible fibrosis. An orthogonal cross-section as shown in Figure 7.13(E) confirms the presence of epithelial BE glands. Some moderate disruption of layered architecture is present, suggesting the presence of LGD. Figure 7.14 shows an enlarged view of the ROI illustrated in Figure 7.13(B,C). This ROI covers the BE region and the edge of the EMR scar. Crosssectional images in the ROI reveal ovular BE glands confined to the mucosa with some distortion of layered architecture. 7.7.2 Comparisonto Histology Figure 7.15 shows a comparison of 3D-OCT data to conventional white light video endoscopy and histology analysis. Figure 7.15(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Figure 7.13. The EMR treatment site presents with raised borders and 144 some irregular texture but cannot be evaluated visually for dysplasia. Figure 7.15(B) shows histology of the distal border of the EMR treatment site. Excellent correlation is apparent between the histology image and the 3D-OCT image in Figure 7.15(C). Small BE glands are visible in both the histology and 3D-OCT images, along with some loss of layered structure. This patient was diagnosed with BE and LGD after analysis of the histology image, consistent with the classification under 3D-OCT. The lateral EMR margin was also positive for BE and LGD under histology. The proximal border was normal under histology, consistent with the lack of BE glands shown in the enface OCT image in Figure 7.13(C). In total, 2 out of the 2 nodule post-EMR patients (100%) showed evidence of BE and LGD under 3DOCT imaging (confirmed with histology for both patients). While a larger study is required to accurately determine the ability of 3D-OCT to evaluate EMR treatment sites for disease recurrence or incomplete resection, these initial results are encouraging. 145 7.8 Figures Figure 7.1. 3D-OCT images of normal squamous mucosa in the esophagus. A: Volume rendering showing location of orthogonal cut planes. B: En face XZ image of epithelium. Dashed box indicates region of interest shown in Figure 3.3. C: YZ Cross-section. D: XY cross-section. Coloured arrows in B-D indicate locations of other orthogonal cut planes. 146 Figure 7.2. Enlarged views of normal esophageal squamous mucosa in the region of interest indicated in Figure 8.1(B). A-D: En face images at tissue depths corresponding approximately to the epithelium (A), lamina propria (B), muscularis mucosa (C), and submucosa (D). Vascular features are visible in the lamina propria (red triangle) and submucosa (black triangle). The same features can be identified in cross-sectional images (E-G). 147 0 EP o LP MM- SM , MP Figure 7.3. Comparison of 3D-OCT data to conventional examination of normal esophageal mucosa. A: White light video endoscopy image of mid-esophagus. B: Representative histology slide from excisional biopsy specimen. C: Cross-sectional OCT image extracted from 3D dataset. EP, epithelium. LP, lamina propria. MM, muscularis mucosa. SM, submucosa. MP, muscularis propria. 148 I GA I L NS I BE Figure 7.4. 3D-OCT images of Barrett's esophagus islands near the GE junction. A: Volume rendering showing location of orthogonal cut planes. B: En face XZ image at 175 um tissue depth. Dashed box indicates region of interest shown in Figure 3.6. C: En face XZ image at 345 um tissue depth. Irregular glandular features are apparent. D: YZ Cross-section. Gastric (GA), normal squamous (NS), and BE regions can be distinguished. E: XZ cross-section showing gastric and transitionary tissue. F: XZ crosssection showing irregular BE glands. Coloured arrows in B-F indicate locations of other orthogonal cut planes. 149 Figure 7.5. Enlarged views of BE island in the region of interest indicated in Figure 8.4(B,C). A: En face image showing transition from normal esophagus to BE. B: YZ cross-sectional image illustrating transition to BE. C: XZ cross-sectional image showing superficial glandular structure and distortion of mucosal layers. Figure 7.6. Comparison of 3D-OCT data to conventional examination of BE island. A: White light video endoscopy image showing BE islands (IS) B: histology slide from excisional biopsy specimen showing BE glands beneath superficial tissue (BE). C: Cross-sectional OCT image extracted from 3D dataset. BE glands are visible adjacent to normal squamous epithelium. 150 Figure 7.7. 3D-OCT images of region previously treated with RFA due to BE. A: Volume rendering showing location of orthogonal cut planes. B: En face XZ image at 100 um tissue depth. Dashed box indicates region of interest shown in Figure 3.9. C: En face XZ image at 390 um tissue depth. Scattered, buried glandular features are apparent. D: YZ Cross-section. Gastric (GA) and normal squamous (NS) regions can be distinguished. Scattered glands possibly representing buried BE (?BE) are buried 350 400 um beneath tissue surface. E: XZ cross-section showing scattered glands (G). Coloured arrows in BE indicate locations of other two orthogonal cut planes. 151 Figure 7.8. Enlarged views of possible buried BE glands (?BE) in region of interest indicated in Figure 8.7(B,C). A: En face image showing ovular lumen shape. B: YZ cross-sectional image. C: XZ crosssectional image. Hyperintense striations in the lamina propria and an isolated gland are visible. Figure 7.9. Comparison of 3D-OCT data to conventional examination of RFA-treated region. A: White light video endoscopy image. GE junction is extended proximally at site of prior RFA (triangle). B: histology slide from excisional biopsy specimen near GE junction. C: Cross-sectional OCT image extracted from 3D dataset. Small glands, possibly scattered buried BE glands (?BE), are visible buried within normal squamous epithelium. 152 I GA II NO J I NS I Figure 7.10. 3D-OCT images of a nodule near the GE junction. A: Volume rendering showing location of orthogonal cut planes. B: En face XZ image at 100 um tissue depth. Dashed box indicates region of interest shown in Figure 3.12. C: En face XZ image at 600 um tissue depth. Region of large, branching glands and vessels is apparent. D: YZ Cross-section. Gastric (GA), nodule (NO), and normal squamous (NS) regions can be distinguished. Large branching glands (BG) extend to the tissue surface. E: XZ cross-section showing highly irregular branching glands (BG). Coloured arrows in B-E indicate locations of other two orthogonal cut planes. 153 Figure 7.11. Enlarged views of an esophageal nodule in the region of interest indicated in Figure 8.10(B,C). A: En face image showing nodule region. B: XZ cross-sectional image showing irregular, branching glandular structure (BG) and disruption of mucosal layers. C: YZ cross-sectional image. Figure 7.12. Comparison of 3D-OCT data to conventional examination of an esophageal nodule. A: White light video endoscopy image. Glandular nodule is visible (triangle). B: histology slide from excisional biopsy specimen. C: Cross-sectional OCT image extracted from 3D dataset. Large, irregular branching glands (BG) are visible. 154 I GA I I I I A: Figure 7.13. 3D-OCT images of region previously treated with EMR to remove an esophageal nodule. tissue um 175 at image XZ face En B: planes. cut orthogonal of location Volume rendering showing depth. Dashed box indicates region of interest shown in Figure 8.15. C: En face XZ image at 350 um tissue depth. Small concentration of BE (BE) glands is apparent, and outline of EMR site can be distinguished (triangles) D: YZ Cross-section. Gastric (GA), BE region (BE), EMR scar (SC) and normal squamous (NS) regions can be distinguished. E: XZ cross-section showing BE glands (BE). Coloured arrows in B-E indicate locations of other orthogonal cut planes. 155 in Figure 8.13(B,C). Figure 7.14. Enlarged views of BE zone following EMR in region of interest indicated BE glands. B: XZ of group A: En face image showing BE region. B: YZ cross-sectional image showing cross-sectional image showing three BE glands. site. Figure 7.15. Comparison of 3D-OCT data to conventional examination of distal margin of prior EMR histology B: (triangle). visible is A: White light video endoscopy image. Prior EMR site with raised margins BE slide from excisional biopsy specimen. C: Cross-sectional OCT image extracted from 3D dataset. Six glands (BE) are visible. 156 7.9 References [1] N. J. Shaheen, "Advances in Barrett's esophagus and esophageal adenocarcinoma," Gastroenterology,vol. 128, pp. 1554-66, May 2005. [2] A. Jemal, R. Siegel, E. Ward, T. Murray, J. Xu, and M. J. Thun, "Cancer statistics, 2007," CA CancerJ Clin, vol. 57, pp. 43-66, Jan-Feb 2007. [3] B. F. Overholt, M. Panjehpour, and J. M. Haydek, "Photodynamic therapy for Barrett's esophagus: follow-up in 100 patients," GastrointestEndosc, vol. 49, pp. 1-7, Jan 1999. [4] X. Chen and C. S. Yang, "Esophageal adenocarcinoma: a review and perspectives on the mechanism of carcinogenesis and chemoprevention," Carcinogenesis,vol. 22, pp. 1119-29, Aug 2001. [5] A. J. Cameron and H. A. Carpenter, "Barrett's esophagus, high-grade dysplasia, and early adenocarcinoma: a pathological study," Am J Gastroenterol,vol. 92, pp. 586-91, Apr 1997. [6] S. J. Spechler, "Screening and surveillance for complications related to gastroesophageal reflux disease," Am J Med, vol. 111 Suppl 8A, pp. 130S-136S, Dec 3 2001. [7] H. Mashimo, M. S. Wagh, and R. K. Goyal, "Surveillance and screening for Barrett esophagus and adenocarcinoma," J Clin Gastroenterol,vol. 39, pp. S33-41, Apr 2005. [8] J. J. Nigro, J. A. Hagen, T. R. DeMeester, S. R. DeMeester, J. Theisen, J. H. Peters, and M. Kiyabu, "Occult esophageal adenocarcinoma: extent of disease and implications for effective therapy," Ann Surg, vol. 230, pp. 433-8; discussion 438-40, Sep 1999. [9] M. S. Dar, J. R. Goldblum, T. W. Rice, and G. W. Falk, "Can extent of high grade dysplasia in Barrett's oesophagus predict the presence of adenocarcinoma at oesophagectomy?," Gut, vol. 52, pp. 486-9, Apr 2003. [10] J. M. Collard, "High-grade dysplasia in Barrett's esophagus. The case for esophagectomy," Chest Surg Clin NAm, vol. 12, pp. 77-92, Feb 2002. [11] G. W. Falk, T. W. Rice, J. R. Goldblum, and J. E. Richter, "Jumbo biopsy forceps protocol still misses unsuspected cancer in Barrett's esophagus with high-grade dysplasia," Gastrointest Endosc, vol. 49, pp. 170-6, Feb 1999. [12] B. J. Dunkin, J. Martinez, P. A. Bejarano, C. D. Smith, K. Chang, A. S. Livingstone, and W. S. Melvin, "Thin-layer ablation of human esophageal epithelium using a bipolar radiofrequency balloon device," Surgical Endoscopy and Other Interventional Techniques, vol. 20, pp. 125-130, Jan 2006. 157 [13] H. Mashimo, Y. Chen, S. W. Huang, Q. Huang, A. Aguirre, J. Schmitt, and J. Fujimoto, "Endoscopic optical coherence tomography reveals Barrett's underneath squarnous neoepithelium after radiofrequency ablation," Gastroenterology,vol. 132, pp. A96-A96, Apr 2007. [14] J. P. Byrne, G. R. Armstrong, and S. E. Attwood, "Restoration of the normal squamous lining in Barrett's esophagus by argon beam plasma coagulation," Am J Gastroenterol,vol. 93, pp. 1810-5, Oct 1998. [15] C. J. Kelty, R. Ackroyd, N. J. Brown, T. J. Stephenson, C. J. Stoddard, and M. W. Reed, "Endoscopic ablation of Barrett's oesophagus: a randomized-controlled trial of photodynamic therapy vs. argon plasma coagulation," Aliment PharmacolTher, vol. 20, pp. 1289-96, Dec 2004. [16] J. L. Van Laethem, M. O. Peny, I. Salmon, M. Cremer, and J. Deviere, "Intramucosal adenocarcinoma arising under squamous re-epithelialisation of Barrett's oesophagus," Gut, vol. 46, pp. 574-7, Apr 2000. [17] K. Ragunath, N. Krasner, V. S. Raman, M. T. Haqqani, C. J. Phillips, and I. Cheung, "Endoscopic ablation of dysplastic Barrett's oesophagus comparing argon plasma coagulation and photodynamic therapy: a randomized prospective trial assessing efficacy and cost-effectiveness," ScandJ Gastroenterol,vol. 40, pp. 750-8, Jul 2005. [18] R. Kiesslich, M. von Bergh, M. Hahn, G. Hermann, and M. Jung, "Chromoendoscopy with indigocarmine improves the detection of adenomatous and nonadenomatous lesions in the colon," Endoscopy, vol. 33, pp. 1001-1006, Dec 2001. [19] D. C. Adler, Y. Chen, R. Huber, J. Schmitt, J. Connolly, and J. G. Fujimoto, "Three-dimensional endomicroscopy using optical coherence tomography," Nature Photonics, vol. 1, pp. 709-716, Dec 2007. [20] Y. Chen, A. D. Aguirre, P. L. Hsiung, S. Desai, P. R. Herz, M. Pedrosa, Q. Huang, M. Figueiredo, S. W. Huang, A. Koski, J. M. Schmitt, J. G. Fujimoto, and H. Mashimo, "Ultrahigh resolution optical coherence tomography of Barrett's esophagus: preliminary descriptive clinical study correlating images with histology," Endoscopy, vol. 39, pp. 599-605, Jul 2007. 158 CHAPTER 8 8.0 Clinical 3D-OCT in the Lower Gastrointestinal Tract 159 8.1 Motivation Chapter 7 described initial clinical study results focusing on the use of 3D-OCT endomicroscopy for pre- and post-treatment assessment of Barrett's esophagus (BE) and esophageal nodules arising from BE. Similar applications for 3D-OCT can also be developed in the lower gastrointestinal (GI) tract. Colorectal cancer is another common GI disease with high morbidity and mortality rates. CRC is the third leading cause of cancer death, accounting for about 10% of cancer deaths overall [1]. Despite its high incidence, colorectal cancer is one of the most detectable, and, if found early, most treatable forms of cancer. Although most colorectal cancers arise from adenomatous polyps that are detectable using conventional endoscopy, many flat (non-polypoid) lesions are missed during routine exams [2]. Up to 50% of these more subtle lesions are missed by conventional endoscopy [3]. Detection of early-stage cancer is particularly difficult in patients with inflammatory bowel diseases (IBD) such as ulcerative colitis (UC) and Crohn's disease (CD), where neoplastic tissue is often flat rather than polypoid in form and multifocal in distribution [4]. As many as 1.4 million individuals in the United States have inflammatory bowel disease [5] and are at increased risk for the development of colorectal cancer [6]. UC in particular affects up to 780,000 individuals in the United States and Canada and is newly diagnosed in 7000 - 46,000 individuals per year [5]. UC is associated with a -5x increase in risk of developing colorectal cancer compared to the general population, with colorectal cancer accounting for one sixth of all deaths in UC patients [6]. Unfortunately, early-stage dysplastic lesions are often flat, diffuse and multifocal in these individuals [4]. As a result, dysplastic lesions are easily obscured by the gross inflammatory background of UC, making early detection extremely challenging. The standard of care for inflammatory bowel disease is similar to BE, with random biopsies being performed throughout the length of the affected region. Colonoscopy with biopsy is generally performed when the disease is in remission to reduce the risk of complications. 3D-OCT, on the other hand, could be used to visualize cross-sectional tissue microstructure without biopsy and with practically no added risk of bleeding in IBD patients. 3D-OCT could also potentially be used to differentiate early dysplastic progression from inflammation based on variations in architectural distortion. As in BE, image guided biopsy and therapy for IBD patients using 3D-OCT promises to improve diagnostic and therapeutic procedures by providing comprehensive 3D tissue visualization. Radiation proctitis (RP) is another form of chronic inflammation and injury of the lower GI tract. This condition is a common side effect of radiation therapy used to treat prostate cancer and occurs in some form in approximately 10 - 15% of patients. RP causes lower GI bleeding and is generally treated with electrocautery or argon plasma coagulation. Healing from these deeply-penetrating ablative methods can be problematic in the area of prior RP, with a high likelihood of ulcerations, structuring, or further 160 bleeding. More recently, radiofrequency ablation (RFA) has been studied as a novel treatment for RP. RFA produces a superficial ablation field covering several square centimeters and has been demonstrated to be effective in treating BE as well as RP in several pilot studies. Since excisional biopsy is contraindicated in RP patients due to an increased risk of bleeding, conventional endoscopy is the primary modality used to assess the disease and the relevant interventional therapies. 3D-OCT can be applied in RP patients to provide subsurface tissue views that are not available with conventional endoscopy. Subsurface information can be used to analyze the extent of RP, plan interventions based on this data, evaluate the healing process on follow-up, and check for recurrence over time. In this section of the thesis work, the same 3D-OCT endomicroscopy system and spiral-scanning imaging catheters used during upper endoscopy were used to assess four tissue categories in the lower GI tract: 1) no pathology ("Normal"); 2) inflammatory bowel diseases ("IBD"); 3) radiation proctitis prior to undergoing RFA therapy ("RP pre-RFA"); and 4) sites previously treated with RFA to control RP ("RP post-RFA"). 3D-OCT was used to establish characteristic features of colonic, rectal, and anal mucosa for each tissue group. 3D-OCT was found to be particularly valuable for pre-treatment and follow-up assessment of RP patients due to the contraindication of excisional biopsy. 8.2 Clinical Protocol All subjects in this study were imaged at the Veterans Affairs Boston Healthcare System (VABHS) Jamaica Plain campus in collaboration with Dr. Hiroshi Mashimo, MD PhD, and Dr. Qin Huang, MD. Subjects were recruited from the pool of patients undergoing elective colonoscopy for screening, surveillance, or treatment of IBD or RP. Subjects followed a standard preparation procedure for colonoscopy, including cessation of blood thinners, fasting, and colonic clearing prior to the appointment. 3D-OCT imaging was performed in tandem with standard video endoscopy. The imaging protocol was very similar to the procedure described in Chapter 7, Section 7.2. For normal subjects, the endoscopist identified regions for 3D-OCT imaging that appeared normal and unremarkable using white light video endoscopy. Imaging was performed by inserting the 3D-OCT catheter down the working channel of a standard endoscope. For IBD patients the endoscopist identified regions of active or controlled disease for 3D-OCT imaging using visual markers such as inflammation, edema, and fibrosis. For RP pre-RFA subjects the endoscopist identified locations suspicious for RP using similar criteria in patients with a history of prostate cancer and previous radiation therapy. For RP post-RFA subjects, the endoscopist identified regions of prior treatment using patient history notes describing the extent and location of the treated region relative to the anal verge. In some cases subtle visual cues could also be used to differentiate treated tissue from untreated tissue. Where clinically 161 indicated, pinch biopsy samples were obtained from the 3D-OCT imaging sites following the conclusion of all image acquisition. Between 3-10 sites were assessed for each subject, increasing the total procedure time by an average of 11 minutes and by no longer than 20 minutes. This increase in procedure length is similar to the time increase associated with chromoendoscopy [3]. After the acquisition of all 3D-OCT datasets, the imaging catheter was withdrawn from the working channel and replaced with pinch biopsy forceps to acquire biopsies of the imaged tissue where clinically indicated. Biopsies were not obtained from normal subjects due to the slight risk associated with pinch biopsy and the lack of clinical necessity. Biopsies were also not obtained from RP patients due to a significant increase in the risk of bleeding. Histopathology from pinch biopsies was evaluated by an expert pathologist in order to establish a tissue classification for each site. The histological diagnoses were then compared to tissue features based on the 3D-OCT data where applicable. Imaging parameters were identical to those described in Chapter 7, Section 7.2. In total, 75 sites from 15 unique patients were imaged during this study from February to December 2008. Table 8.1 shows a summary of the collected 3D-OCT image data. Note that the total number of patients in Table 8.1 exceeds 15 since multiple tissue types were imaged from several patients, typically consisting of a normal region and a pathological region. The "other" grouping includes pathologic tissue that was outside the main scope of the study and primarily consisted of ulcers and polyps. The following sections show representative 3D-OCT data, characteristic features, and comparisons to histology for each main tissue type. Tissue Type Normal Colon IBD RP Pre-RFA RP Post-RFA Other Pathology Number of Patients 12 1 2 3 8 Number of Imaging Sites 28 2 3 11 31 Table 8.1. Summary of 3D-OCT data collected during lower endoscopy. 8.3 Normal Colon Obtaining normal image data was a necessary first step in this study in order to establish a baseline for comparison to pathologic tissue. 3D-OCT images of normal colonic mucosa were obtained at 28 sites from 12 unique patients. All patients tolerated the procedure well and there were no known immediate or long-term complications. Useable data was obtained for 10 out of the 12 patients. 2 patients produced unusable data due to nonuniform probe rotation caused by excessive pressure on the colonoscope. Results of the normal colon imaging are described in Sections 8.3.1 to 8.3.2 below. 162 8.3.1 CharacteristicFeatures Figure 8.1 shows typical 3D-OCT data obtained from the normal colon. Figure 8.1(A) shows a volumetric rendering of the entire 980 x 512 x 1268 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 7.5 mm, and the imaging range was 1.7 mm in tissue. The images in Figure 8.1 have been axially cropped to remove the probe sheath and non-useable points beyond the penetration depth of light in the tissue. The useable imaging depth in tissue was approximately 1100 jtm. Unless otherwise specified, all crosssectional and en face images shown below are formed by averaging the volumetric data over a 20 jtmthick section to reduce speckle noise. Since 20 jtm is approximately the size of a single epithelial cell, minimal image blurring is observed from this process [7]. The red, green, and blue boxes in Figure 8.1(A) illustrate the locations of orthogonal cut planes used to extract enface and cross-sectional image slices for closer examination. In this and all subsequent 3D-OCT image data the X axis represents the probe's rotational axis (fast, 50 - 70 Hz), the Y axis represents the axial OCT line data, and the Z axis represents the probe's pullback direction (slow, 500 - 1000 jtm/s). Figure 8.1(B) shows an en face image centered at a tissue depth of 300 jim, corresponding to the bottom of the columnar epithelial layer. This image is similar in concept to a conventional microscopy image obtained by setting the focus to a depth of 300 jtm. The left side of the image is more proximal colon and the ride side is more distal (anal). Consistent with previous OCT studies on excised pathology lab samples [8] the normal columnar epithelium of the colon shows low light penetration. Colonic crypts are visible as round or slightly oval regions of hyperintensity, thought to be caused by increased light transmission through the crypt lumen [8]. The large hyperintense region at a pullback distance of approximately 3 mm is scar tissue from a prior polypectomy at this site. Transverse probe jitter is reduced in the colon compared to the esophagus due to a significant reduction in respiratory motion and a complete lack of heartbeat motion. Longitudinal (Z-oriented) artifacts are primarily caused by variations in probe pullback velocity due to periodic sticking and slipping within the outer sheath. The blue and green arrow pairs on the border of Figure 8.1(B) indicate the positions of the cross-sectional images shown in Figure 8.1(C,D). Figure 8.1(C) shows a YZ cross-sectional image at the position indicated by the green arrow pair in Figure 8.1 (B). In regions outside of the prior polypectomy site vertical streaks are present due to increase light transmission through crypt lumens. The polypectomy site exhibits a hyperintense layer of fibrotic scar tissue covered by a featureless tissue layer. It is difficult to discern a layered architecture at most locations in the colon due to poor transmission through the columnar epithelium. Figure 8.1(D) shows an 163 XY cross-sectional image at the position indicated by the blue arrow pair in Figure 8.1(B). Similar features to the cross-section in Figure 8.1(C) can be appreciated. The dense spatial sampling density, high optical resolution, and large field of view of 3D-OCT endomicroscopy enables high-magnification analysis of features such as colonic crypts over arbitrary regions of interest (ROI). Crypts are the main glandular structures in the human colon, and changes in crypt size and appearance are associated with the earliest forms of colorectal cancer [9] and other diseases. The ability to assess the 3D structure of crypts, possibly using automated classification algorithms [10], is therefore of potential value for future applications in cancer detection and treatment. Figure 8.2(A) shows an en face images of the ROI shown as a dashed box in Figure 8.1 (B). To generate the clearest view of the crypts, the axial data was averaged over a depth of 480 tm. The average crypt diameter is approximately 75 - 150 tm. Variations in apparent crypt size can be attributed to variations in tissue orientation and some changes in probe pullback velocity due to stress on the probe from the endoscope. Figure 8.2(B,C) shows the expected regular hyperintense streaks associated with the crypt lumens. 8.3.2 Comparison to Histology Comparison of 3D-OCT endomicroscopy images to histology slides obtained from excisional biopsy and to white light video endoscopy examination is a good method for linking image features to known tissue structures. Figure 8.3(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 8.3.1. The colonic mucosa appears pale, smooth, and regular, consistent with normal columnar epithelial mucosa. Figure 8.3(B) shows representative histology of normal colonic glandular mucosa. Since there was no clinical indication to biopsy this particular patient, the histology image in Figure 8.3(B) is taken from another subject where biopsy was required but a normal diagnosis was made by the pathologist. Good correlation is shown between the histology image and the 3D-OCT image. The epithelium, muscularis mucosa, and submucosa are visible under histology and under 3D-OCT as shown in Figure 8.3(C). The lamina propria is not generally distinguishable from the epithelium under 3D-OCT due to its close integration with the epithelial crypts. The muscularis mucosa and submucosa boundary is visible at this particular location but is generally not appreciable. In total, 10 out of the 10 normal patients (100%) with useable 3D-OCT data showed features consistent with those described above. 8.4 Inflammatory Bowel Diseases 3D-OCT images of active UC were obtained at 2 sites from 1 unique patient to demonstrate imaging in an IBD case. The patient tolerated the procedure well and there were no known immediate or long-term 164 complications. This patient had previously undergone a partial colectomy related to his UC, with the remaining colon pulled down and re-attached to the anus and rectum. He presented with endoscopically apparent active UC in a small region near the squamo-columnar junction. The squamo-columnar junction in the colon is sometimes referred to as the "dentate line". Useable data was obtained without significant motion artifacts. Results of the UC imaging are described in Sections 8.4.1 to 8.4.2 below. 8.4.1 CharacteristicFeatures Figure 8.4 shows 3D-OCT data obtained from a region of active UC. Figure 8.4(A) shows a volumetric rendering of the entire 771 x 512 x 1794 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 8.6 mm, and the imaging range was 1.7 mm in tissue. The images in Figure 8.4 have been axially cropped to remove the probe sheath and non-useable points beyond the penetration depth of light in the tissue. The useable imaging depth in tissue was approximately 1100 lm. The red, green, blue and purple boxes in Figure 8.4(A) illustrate the locations of orthogonal cut planes used to extract en face and cross-sectional image slices for closer examination. Figure 8.4(B) shows an en face image centered at a tissue depth of 200 jtm, corresponding to where the columnar epithelial layer should be located in normal mucosal tissue. A complete lack of regular crypt architecture is immediately apparent from inspection of the en face 3D-OCT image. Large subsurface voids and bands of hyperscattering tissue, possibly fibrotic, are clearly visualized. A wedge of comparatively normal squamous epithelial mucosa is visible at the right of the image, demarcating the boundary between pathologic UC and the squamous epithelial mucosa of the anal verge. The blue and green arrow pairs on the border of Figure 8.4(B) indicate the positions of the cross-sectional images shown in Figure 8.4(C,D). Figure 8.4(C) shows a YZ cross-sectional image at the position indicated by the green arrow pair in Figure 8.4(B). UC and normal squamous epithelial mucosa are visible on the left and right sides of the image respectively. The large -1.5 mm subsurface void near the centre of Figure 9.4(C) may be an ulcerative tunnel formed by repeated epithelial stripping and regeneration during disease remission. A large number of superficial vessels and edematous regions are present in the UC section, along with a lack of regular squamous or columnar epithelial architecture. Figure 8.4(D) shows an XZ cross-sectional image at the position indicated by the blue arrow pair in Figure 8.4(B). Similar ulcerative features are visible here as well. Figure 8.5 and Figure 8.6 show detailed views of the two ROI's indicated in Figure 8.4(B). Figure 9.5(A) shows an en face view of an ulcerative region with tortuous and irregular superficial vasculature. The cross-sectional images in Figure 8.5(B,C) show details of the UC architecture. There is a stark 165 contrast between this region and the region shown in Figure 8.6. The en face view in Figure 8.6(A) at the same tissue depth is featureless and unremarkable, consistent with squamous epithelial mucosa. This is confirmed with the cross-sectional images in Figure 8.5(B,C). The tissue shown in Figure 9.5(B,C) contains subtle hyperintense striations in the epithelium not found in the squamous epithelial mucosa of completely healthy subjects. These horizontal bands of tissue may be fibrotic structures resulting from healing during disease remission. 8.4.2 Comparison to Histology Figure 8.7(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Figure 8.4. The mucosa appears red, inflamed and ulcerative. The 3D-OCT catheter is also shown in position prior to imaging. Figure 8.7(B) shows representative histology of UC in glandular mucosa. Since there was an increased risk of bleeding in this particular patient, the histology image in Figure 8.7(B) is taken from another subject with a similar case of active UC. Lymphocytic mucosal infiltration is present along with submucosal fibrosis. Ulceration results in the formation of a pseudopolyp as the epithelium is stripped away to expose the submucosa. Good correlation is shown between the histology image and the 3D-OCT image. Large edematous regions, loss of normal crypt patterns, and disruption of the epithelium are apparent under 3D-OCT and histology. While more study is required to expand on these findings, 3D-OCT could potentially be useful in assessing the severity of IBD, monitoring response to therapy, and differentiating dysplastic lesions from the gross inflammatory background. 8.5 Radiation Proctitis: Pre-Radiofrequency Ablation 3D-OCT images of RP prior to receiving RFA were obtained at 3 sites from 2 unique patients. This data was collected in order to establish characteristics of active, untreated RP and forms a baseline for comparison of treated RP and for assessing the effects of RFA. Both of these patients had not previously been treated with RFA. 1 patient had previously undergone argon plasma coagulation to control the RP but had experienced additional bleeding subsequent to treatment. Both patients tolerated the 3D-OCT imaging procedure well and there were no known immediate or long-term complications. Useable data was obtained for 2 out of the 2 patients. Results of the RP pre-RFA imaging are described in Sections 8.5.1 to 8.5.2 below. 8.5.1 CharacteristicFeatures Figure 8.8 shows typical 3D-OCT data obtained from an ectatic RP region prior to any endoscopic or surgical therapy. This patient presented with rectal bleeding following 8 weeks of radiation therapy for 166 prostate cancer. The patient completed radiation therapy in mid-2001 and was imaged with 3D-OCT in 2008. It is not uncommon for RP to develop many years after completion of a radiation course. Endoscopic examination revealed irregular and ectactic blood vessels along with mucosal inflammation in the rectum consistent with RP. In this case the goal of 3D-OCT imaging was not to assist in making a diagnosis of RP, but rather to obtain subsurface tissue information for pathology characterization. Since excisional biopsies cannot be performed in RP patients due to a significantly increased risk of bleeding, 3D-OCT is one of the only ways to obtain such information. Figure 8.8(A) shows a volumetric rendering of the entire 1202 x 512 x 1240 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 9.0 mm, and the imaging range was 1.7 mm in tissue. The red, green and blue boxes in Figure 8.8(A) illustrate the locations of orthogonal cut planes used to extract enface and cross-sectional image slices for closer examination. Figure 8.8(B) shows an en face image centered at a tissue depth of 250 pim, corresponding to a depth slightly lower than the centre of the columnar epithelial layer. A regular crypt pattern is visible over most of the tissue, although transverse probe jitter caused by nonuniform rotation partially obscures the features. Several large vessels can be seen near the top right of the image (triangles). Vessels of this diameter (-500 pm) are not normally found at superficial depths and are indicative of ectasia arising from RP. The blue and green arrow pairs on the border of Figure 8.8(B) indicate the positions of the crosssectional images shown in Figure 8.8(C,D). Figure 8.8(C,D) shows cross-sectional images bisecting the irregular vessels identified in Figure 8.8(B). The ability to align cross-sectional image planes to surface features such as ectatic vessels is a significant advantage of 3D-OCT. In Figure 8.8(C,D), the vessels are visible as shadowing and slightly hypointense structures located in the epithelial layer (triangles). The rest of the tissue appears fairly normal and exhibits typical hyperintense streaking associated with colonic crypts. Figure 8.9 shows an enlarged view of the ROI illustrated in Figure 8.8(B). This ROI spans a superficial vessel that is associated with RP. The vessel is clearly identifiable in the en face image as a large, irregular hypointense structure (triangles). Cross-sectional images in the ROI highlight the extremely superficial nature of the vessels. The overlying tissue is less than 150 pm thick, suggesting a high likelihood of bleeding resulting from slight agitation such as stool passage. Vessels of this large size and superficial nature are not seen in the squamous epithelial mucosa of normal patients. In regions immediately adjacent to the vessel there is a lack of normal colonic crypts and presence of a quasi-regular layered structure, indicating that the vessels are embedded under the squamous epithelial mucosa of the anal verge. 167 8.5.2 Comparison to Histology Figure 8.10 shows a comparison of 3D-OCT data to conventional white light video endoscopy and histology analysis. Figure 8.10(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 8.5.1. Excellent correlation is shown between the histology image in Figure 8.10(B) and the 3D-OCT image in Figure 8.10(C). The colonic mucosa contains inflammatory infiltrates and large superficial vessels with a disrupted crypt pattern. In total, 2 of the 2 RP pre-RFA patients (100%) showed features consistent with those described above. A significant degree of variability was observed in vessel size and orientation, although in both cases the vessels were superficial and irregular. The ability to obtain enface and cross-sectional tissue architecture without excisional biopsy is a major advantage for RP patients, enabling baseline tissue characterization for later analysis of treatment effects. 8.6 Radiation Proctitis: Post-Radiofrequency Ablation 3D-OCT images of RP subsequent to receiving RFA were obtained at 11 sites from 2 unique patients. This data was compared to pre-treatment images to study the effects of RFA for treating RP, which is a less-established application than treatment of BE. One patient was imaged 12 months and 14 months after receiving RFA therapy, and in both cases useable 3D-OCT data was obtained. The other patient was imaged within 10 minutes of receiving RFA. This did not produce useable 3D-OCT data due to the presence of tissue debris and blood in the imaging field resulting from the application of RFA. Both patients tolerated the procedure well and there were no known immediate or long-term complications. Results of the RP post-RFA imaging are described in Sections 8.6.1 to 8.6.2 below. 8.6.1 CharacteristicFeatures Figure 8.11 shows typical 3D-OCT data obtained from a formerly ectatic RP region that was treated with RFA 14 months prior to imaging. This patient presented with no signs or symptoms of active bleeding or proctitis since receiving RFA. Endoscopic examination revealed regular columnar epithelial mucosa in the colon and smooth squamous epithelial mucosa in the rectum consistent with resolution of the prior RP. No large ectatic vessels or overt inflammation were detected under white light video endoscopy. Figure 8.11(A) shows a volumetric rendering of the entire 771 x 512 x 1794 (lines x axial pixels x frames, X x Y x Z) dataset, which was acquired in -21 s. The pullback length was 20.7 mm, the rotational circumference was 7.2 mm, and the imaging range was 1.7 mm in tissue. The red, green and blue boxes in Figure 8.11(A) illustrate the locations of orthogonal cut planes used to extract en face and cross-sectional image slices. 168 Figure 8.11(B) shows an enface image centered at a tissue depth of 350 tpm, corresponding to a depth slightly beneath the epithelial layer. The probe was pulled back over the dentate line and a clear demarcation between the proximal columnar epithelial mucosa and the distal (anal) squamous epithelial mucosa is apparent. A regular crypt pattern is visible over the region containing columnar epithelium on the left side of the image. The large hypointense structure at the far left is a hemorrhoid as noted during endoscopic examination of the region. The squamous epithelial mucosa at this depth, likely corresponding to lamina propria, is hyperintense and featureless with the exception of several large vascular structures (triangles). These structures are very similar to the ectatic vessels visualized prior to RFA treatment in Figure 8.8, but at 14 months post-RFA they are now found at a deeper tissue location. The blue and green arrow pairs on the border of Figure 8.11(B) indicate the positions of the cross-sectional images shown in Figure 8.11(C,D). Figure 8.11 (C,D) shows cross-sectional images bisecting the vessels identified in Figure 8.11(B). The combination of en face and cross-sectional data enables rapid identification and localization of vascular structures as well as 3D architectural analysis. In Figure 8.11(C,D), the vessels are visible as nonshadowing and slightly hypointense structures (triangles) located beneath the epithelial and lamina propria layers of regular-appearing squamous epithelial mucosa. The rest of the tissue appears fairly normal and exhibits typical layered architecture consistent with squamous epithelial mucosa. In comparison to the pre-treatment images shown in Figure 8.8, the vessels exposed to RFA are deeply buried and show a reduction in axial shadowing that may coincide with reduced blood flow from reduced vessel diameter, although this conclusion is speculative at this point. Figure 8.12 shows an enlarged view of the ROI illustrated in Figure 8.11(B). This ROI spans a buried vessel that is associated with a previously ectatic vessel that had been treated with RFA. Again, the vessel is clearly identifiable in the en face image as a large, irregular hypointense structure (triangles). Crosssectional images in the ROI highlight the difference in tissue architecture as a result of RFA. The vessel is covered in a thick, protective layer of squamous epithelium and lamina propria approximately 300 iim thick. The untreated vessels, by comparison, were covered in a thin tissue cap less than 150 Rm thick. The presence of squamous epithelium over top of the vessels could be a result of neo-squamous growth in a region previously occupied by glandular columnar epithelial mucosa. Neo-squamous re-epithelialization has been observed using 2D-OCT in the study of RFA for BE treatment, and a similar process could also occur in the colon. 8.6.2 Comparison to Histology Figure 8.13 shows a comparison of 3D-OCT data to conventional white light video endoscopy. No histology data was available for the post-RFA subjects due to the possibility of incomplete treatment and 169 therefore an elevated risk of bleeding. Additionally, the squamous epithelial mucosa of the anal verge is innervated and therefore pinch biopsy is a painful procedure that can only be performed when there is clinically need. Figure 8.13(A) shows an endoscopic video capture obtained prior to 3D-OCT imaging of the same patient presented in Section 8.6.1. The dentate line is clearly visible and there are no obvious signs of edema, ectatic vessels, or large-scale inflammation. The 3D-OCT image in Figure 8.13(B) shows another buried vessel beneath the lamina propria. In the future, modifications to the study protocol to allow excisional biopsy following successful RFA treatment would be valuable for confirming the presence of deeply-buried vascular remnants in these subjects. Alternatively, if the increased risk of bleeding is deemed to be unacceptably high, animal models could be used to obtain histological correlates of pre- and post-RFA RP sites. In this brief pilot study, 3D-OCT has shown great potential for understanding the architectural morphology associated with RP and for evaluating the mechanism of action of novel RFA therapy. 170 8.7 Figures Figure 8.1. 3D-OCT images of normal columnar epithelial mucosa in the colon. A: Volume rendering showing location of orthogonal cut planes. B: En face XZ image of epithelium. Dashed box indicates region of interest shown in Figure 9.2. C: YZ Cross-section. D: XY cross-section. Coloured arrows in B-D indicate locations of other orthogonal cut planes. Figure 8.2. Enlarged views of normal colonic crypts in the region of interest indicated in Figure 9.1(B). A: En face image formed by projecting volumetric dataset over a 480 um range. Crypt lumens are clearly visible due to increased light transmission. B,C: Cross-sectional images showing typical columnar structure of crypts with increased transmission (C). Red arrows indicate projection region for forming en face image in A. 171 4- EP P- 4-MM --SM -4 of normal glandular mucosa in the Figure 8.3. Comparison of 3D-OCT data to conventional examination histology slide from excisional biopsy colon. A: White light video endoscopy image. B: Representative Crypts (C) are visible but deeper specimen. C: Cross-sectional OCT image extracted from 3D dataset. layers cannot be distinguished. location of orthogonal cut Figure 8.4. 3D-OCT images of ulcerative colitis. A: Volume rendering showing of interest shown in regions indicates boxes planes. B: En face XZ image at 300 um tissue depth. Dashed (UC) and Ulcerative Cross-section. YZ C: apparent. Figure 3.5. Irregular vascular and cystic features are and vessels surface showing cross-section XZ D: normal squamous (NS) regions can be distinguished. planes. cut orthogonal other of locations edema. Coloured arrows in B-D indicate 172 of interest indicated in Figure 9.4(B). Figure 8.5. Enlarged views of ulcerative colitis region in first region vessels and edema. B,C: Cross A: En face image showing irregular subsurface structure including sectional images showing region of ulcerative colitis. of interest indicated in Figure 8.6. Enlarged views of squamous epithelial mucosa in second region mucosa. B,C: Cross Figure 9.4(B). A: En face image showing regular, unremarkable squamous epithelial possibly due to sectional images showing regular layered structure with subtle horizontal striations, healing. 173 A: White light Figure 8.7. Comparison of 3D-OCT data to conventional examination of ulcerative colitis. C: Crossspecimen. biopsy excisional from slide histology video endoscopy image. B: Representative architecture layered of loss and (ED) regions Edematous dataset. 3D from sectional OCT image extracted are apparent. Figure 8.8. 3D-OCT images of region containing edematous vessels due to radiation proctitis. A: Volume depth. rendering showing location of orthogonal cut planes. B: En face XZ image at 250 um tissue located vessels Large Dashed box indicates region of interest shown in Figure 9.9. C: YZ Cross-section. 100 - 150 um below the tissue surface can be visualized (triangles). D: XZ cross-section showing multiple superficial vessels. 174 in Figure 9.8(B). A: En Figure 8.9. Enlarged views of ectatic vessels in the region of interest indicated showing detail of images face image showing large edematous vessel (triangles). B,C: Cross-sectional superficial vessels with thin, hyperintense tissue cap. from radiation Figure 8.10. Comparison of 3D-OCT data to conventional examination of ectatic vessels specimen. C: biopsy proctitis. A: White light video endoscopy image. B: histology slide from excisional are visible in (triangles) vessels Cross-sectional OCT image extracted from 3D dataset. Large superficial all three images. 175 I NC II NS location of Figure 8.11. 3D-OCT images of region treated with RFA for RP. A: Volume rendering showing region of indicates box orthogonal cut planes. B: En face XZ image at 350 um tissue depth. Dashed are now (triangles) interest shown in Figure 3.9. C: YZ Cross-section. Previously edematous vessels vessels. covered in a protective neo-squamous layer. D: XZ cross-section showing covered Figure 8.12. Enlarged views of treated radiation proctitis in the region of interest indicated in Figure 300 3.11(B). A: En face image showing covered vessel (triangles). B,C: Cross-sectional images showing vessels. edematous - 350 um thick protective layer of neo-squamous epithelium over top of previously 176 A: White light video endoscopy Figure 8.13. Comparison of 3D-OCT data to conventional examination. OCT image extracted from image. B: histology slide from excisional biopsy specimen. C: Cross-sectional squamous epithelium. appearing 3D dataset. Buried vessels (triangle) are visible underneath regular 177 8.8 References [1] A. Jemal, R. Siegel, E. Ward, T. Murray, J. Xu, and M. J. Thun, "Cancer statistics, 2007," CA CancerJ Clin, vol. 57, pp. 43-66, Jan-Feb 2007. [2] D. K. Rex, "Maximizing detection of adenomas and cancers during colonoscopy," American Journalof Gastroenterology,vol. 101, pp. 2866-2877, Dec 2006. [3] R. Kiesslich, M. von Bergh, M. Hahn, G. Hermann, and M. Jung, "Chromoendoscopy with indigocarmine improves the detection of adenomatous and nonadenomatous lesions in the colon," Endoscopy, vol. 33, pp. 1001-1006, Dec 2001. [4] S. H. Itzkowitz and N. Harpaz, "Diagnosis and management of dysplasia in patients with inflammatory bowel diseases," Gastroenterology,vol. 126, pp. 1634-1648, May 2004. [5] E. V. Loftus, "Clinical epidemiology of inflammatory bowel disease: Incidence, prevalence, and environmental influences," Gastroenterology,vol. 126, pp. 1504-1517, May 2004. [6] J. A. Eaden, K. R. Abrams, and J. F. Mayberry, "The risk of colorectal cancer in ulcerative colitis: a meta-analysis," Gut, vol. 48, pp. 526-535, Apr 2001. [7] D. C. Adler, Y. Chen, R. Huber, J. Schmitt, J. Connolly, and J. G. Fujimoto, "Three-dimensional endomicroscopy using optical coherence tomography," Nature Photonics,vol. 1, pp. 709-716, Dec 2007. [8] P. L. Hsiung, L. Pantanowitz, A. D. Aguirre, Y. Chen, D. Phatak, T. H. Ko, S. Bourquin, S. J. Schnitt, S. Raza, J. L. Connolly, H. Mashimo, and J. G. Fujimoto, "Ultrahigh-resolution and 3dimensional optical coherence tomography ex vivo imaging of the large and small intestines," GastrointestEndosc, vol. 62, pp. 561-574, Oct 2005. [9] T. Takayama, S. Katsuki, Y. Takahashi, M. Ohi, S. Nojiri, S. Sakamaki, J. Kato, K. Kogawa, H. Miyake, and Y. Niitsu, "Aberrant crypt foci of the colon as precursors of adenoma and cancer," New EnglandJournalof Medicine, vol. 339, pp. 1277-1284, Oct 29 1998. [10] X. Qi, Y. S. Pan, Z. L. Hu, M. V. Sivak, J. Willis, K. Olowe, and A. M. Rollins, "Morphological feature quantification of colonic crypt patterns using microscope-integrated OCT," Gastroenterology,vol. 134, pp. A577-A577, Apr 2008. 178 CHAPTER 9 9.0 Conclusions, Future Work, and Publications 179 9.1 Summary of Thesis Work This thesis project involved a combination of technology development, pre-clinical imaging experiments, and human clinical studies in the field of three-dimensional optical coherence tomography (3D-OCT). First, fundamental advances were made in high-speed wavelength swept light sources to enable the next generation of high-speed 3D-OCT platforms. The Fourier domain mode locked (FDML) laser was proven to be an ideal light source for 3D-OCT due to its unprecedented combination of high sweep rate, wide tuning range, high output power, and low phase noise properties. FDML laser technology was then integrated into a number of 3D-OCT imaging systems. One system was used to demonstrate OCT phase profilometry with sub-nanometer axial resolution at record imaging speeds. This technology could be used in the future for studies of cellular dynamics, industrial inspection of MEMS or microfluidic devices, or other applications requiring measurement of extremely small and extremely rapid transient events. Using a similar phase-sensitive OCT system and FDML laser, a new method was developed to detect gold nanoparticle contrast agents with 1 to 2 orders of magnitude higher signal-to-noise ratios than previously reported OCT methods. This new method used photothermal modulation of the sample and frequency resolved signal detection to reject noise in a manner similar to lock-in detection. This technique could enable molecularly sensitive 3D-OCT imaging of tissue pathologic state by bioconjugation of molecular probes to the nanoshells. This advance would, if successful, fundamentally transform the OCT field by enabling the simultaneous analysis of 3D tissue structure and function in vivo and in real time. A state-of-the-art 3D-OCT endomicroscopy system was constructed in collaboration with LightLab Imaging Inc. in order to perform 3D-OCT imaging of the human gastrointestinal (GI) tract. This system combined an FDML laser with purpose-built data acquisition hardware, signal processing algorithms, and real-time image display. Specialized spiral-scanning imaging probes were developed to provide an optimized combination of flexibility, strength, rotational and pullback uniformity, and optical performance. This system was used in a series of clinical pilot studies including the first demonstration of 3D-OCT imaging in the human colon. New GI applications of 3D-OCT technology focusing on the use of high-speed high-resolution imaging as an adjunct to endoscopic therapy were developed. These studies demonstrated the utility of 3D-OCT for pre- and post-treatment analysis of radiofrequency ablation for both Barrett's esophagus and radiation proctitis. In addition, 3D-OCT was shown to be useful for assessment of excision margins following endoscopic mucosal resection of dysplastic nodules in the esophagus. These studies have opened the door for clinical acceptance of new indications for 3D-OCT, focusing on the targeted imaging of known pathologic regions for treatment planning and follow-up assessment. 180 The core technology developed in this thesis work can also be applied to the more traditional OCT niches of screening and initial detection or diagnosis of pathology. As endoscopic therapies continue to gain in sophistication, capability, and popularity within the gastroenterology community, 3D-OCT can fill an unmet need for providing on-demand, real time 3D visualization of subsurface tissue microstructure both pre- and post-treatment. 9.2 Future Work Future efforts for the projects described in this thesis work should focus on four main areas: 1) In vivo detection of molecularly targeted nanoshells; 2) 3D-OCT imaging engine development; 3) advanced imaging probe design, and; 4) long-term clinical studies. First, the fundamental advances in nanoshell detection described in this thesis work must be translated into successful in vivo imaging experiments. The nanoshells should be bioconjugated to antibodies against proteins overexpressed in common human pathologies. One good example would be an antibody targeting epidermal growth factor receptor (EGFR), which is associated with a number of epithelial cancers including colorectal cancer. These targeted nanoshells could be tested on in vitro cell cultures as well as animal models of colorectal cancer to demonstrate direct detection of pathologic state using 3DOCT. Second, 3D-OCT imaging engine development should integrate the newest advances in high-speed data acquisition technology, hardware-based signal processing, and efficient image display and storage methods to scale up imaging speeds to the rate supported by state-of-the-art FDML laser designs. The first generation 3D-OCT endomicroscopy system developed here was capable at operating at 60 - 100 kHz but could only be run at 60 kHz to ensure optimal image quality for human clinical studies. With currently developed FDML lasers already operating at sweep rates of 370 kHz and with future designs likely to target speeds of 500 kHz or more, the imaging engine must be updated to keep pace with laser technology developments. New data acquisition hardware operating at 400 MSamples/s with onboard digital signal processing chips, combined with recently released multicore personal computer central processing units, could provide a viable pathway for this effort. These increased imaging speeds will enable corresponding increases in spatial sampling density and field of view, providing clinicians with an unsurpassed convergence of true 3D resolution, depth-resolved imaging, and rapid acquisition speeds. Third, new imaging probes must be developed to provide further improvements in image quality and reliability to enable long-term clinical studies of 3D-OCT endomicroscopy. Despite the advances in probe technology described in this thesis work, image quality can still be further improved by reducing nonuniformities in push/pull actuation. This problem may be addressed by further miniaturization of the current design. Entirely new probe concepts, such as forward-imaging raster scan probes, could also be 181 developed to provide extremely rapid acquisition over a smaller field of view. Probe designs must also be scaled to the higher scanning speeds envisioned for the next generation of 3D-OCT imaging engines. Finally, long-term clinical studies must build on the initial results demonstrated in this thesis work to decisively convince the clinical community of the utility of 3D-OCT endomicroscopy. Studies of 3DOCT for pre- and post-therapy analysis can be conducted with long term follow-up of patients treated with radiofrequency ablation, endoscopic mucosal resection, cryospray therapy, photodynamic therapy, or other emerging endoscopic treatments. 3D-OCT could also be studied as an intra-therapy tool for guiding the dosage or resection depth of theses techniques. Studies in 3D-OCT screening for dysplasia in the setting of Barrett's esophagus or inflammatory bowel disease could also be revisited as fields of view become larger and resolutions become higher. Successful clinical studies are the only way for 3D-OCT to become an accepted clinical tool outside of the specialty research field, and are critical for long-term success of the technology. 9.3 Publications Produced During Thesis Work Publications [1-15] were produced between 2006 - 2009 during work on the PhD thesis. Publications [1619] were produced between 2002 - 2004 during work on the SM thesis. [1] D. C. Adler, C. Zhou, T.-H. Tsai, H.-C. Lee, L. Becker, J. M. Schmitt, J. G. Fujimoto, and H. Mashimo, "Three-dimensional optical coherence tomography examination of non-ablated Barrett's esophagus and buried glands beneath neo-squamous epithelium following radiofrequency ablation," Endoscopy, In Preparation, 2009. [2] A. D. Aguirre, D. C. Adler, and J. G. Fujimoto, "Practical Guidelines for Biological Imaging with Optical Coherence Tomography," Nature Protocols, In Preparation, 2009. [3] C. Zhou, D. C. Adler, L. Becker, Y. Chen, T.-H. Tsai, M. Figueiredo, J. Schmitt, J. G. Fujimoto, and H. Mashimo, "Effective treatment of chronic radiation proctitis using radiofrequency ablation," Therapeutic advances in gastroenterology,In Press, 2009. [4] D. C. Adler, C. Zhao, T.-H. Tsai, J. Schmitt, Q. Huang, H. Mashimo, and J. G. Fujimoto, "Threedimensional endomicroscopy of the human colon using optical coherence tomography," Optics Express, vol. 17, pp. 784-796, January 2009 2009. [5] B. R. Biedermann, W. Wieser, C. M. Eigenwillig, G. Palte, D. C. Adler, V. J. Srinivasan, J. G. Fujimoto, and R. Huber, "Real time en face Fourier-domain optical coherence tomography with direct hardware frequency demodulation," Optics Letters, vol. 33, pp. 2556-2558, Nov 1 2008. 182 [6] V. J. Srinivasan, D. C. Adler, Y. Chen, I. Gorczynska, R. Huber, J. Duker, J. S. Schuman, and J. G. Fujimoto, "Ultrahigh-speed Optical Coherence Tomography for Three-Dimensional and En Face Imaging of the Retina and Optic Nerve Head," IOVS, 2008, pp. 08-2127. [7] P. M. Andrews, Y. Chen, M. L. Onozato, S. W. Huang, D. C. Adler, R. A. Huber, J. Jiang, S. E. Barry, A. E.Cable, and J. G. Fujimoto, "High-resolution optical coherence tomography imaging of the living kidney," LaboratoryInvestigation, vol. 88, pp. 441-449, Apr 2008. [8] D. C. Adler, S. W. Huang, R. Huber, and J. G. Fujimoto, "Photothermal detection of gold nanoparticles using phase-sensitive optical coherence tomography," Optics Express, vol. 16, pp. 4376-4393, Mar 31 2008. [9] D. C. Adler, Y. Chen, R. Huber, J. Schmitt, J. Connolly, and J. G. Fujimoto, "Three-dimensional endomicroscopy using optical coherence tomography," Nature Photonics, vol. 1, pp. 709-716, Dec 2007. [10] D. C. Adler, J. Stenger, I. Gorczynska, H. Lie, T. Hensick, R. Spronk, S. Wolohojian, N. Khandekar, J. Y. Jiang, S. Barry, A. E.Cable, R. Huber, and J. G. Fujimoto, "Comparison of three-dimensional optical coherence tomography and high resolution photography for art conservation studies," Optics Express, vol. 15, pp. 15972-15986, Nov 26 2007. [11] R. Huber, D. C. Adler, V. J. Srinivasan, and J. G. Fujimoto, "Fourier domain mode locking at 1050 nm for ultra-high-speed optical coherence tomography of the human retina at 236,000 axial scans per second," Optics Letters, vol. 32, pp. 2049-2051, Jul 2007. [12] S.W. Huang, A. D. Aguirre, R. A. Huber, D. C. Adler, and J. G. Fujimoto, "Swept source optical coherence microscopy using a Fourier domain mode-locked laser," Optics Express, vol. 15, pp. 6210-6217, May 2007. [13] M. W. Jenkins, D. C. Adler, M. Gargesha, R. Huber, F. Rothenberg, J. Belding, M. Watanabe, D. L. Wilson, J. G. Fujimoto, and A. M. Rollins, "Ultrahigh-speed optical coherence tomography imaging and visualization of the embryonic avian heart using a buffered Fourier Domain Mode Locked laser," Optics Express, vol. 15, pp. 6251-6267, May 2007. [14] D. C. Adler, R. Huber, and J. G. Fujimoto, "Phase-sensitive optical coherence tomography at up to 370,000 lines per second using buffered Fourier domain mode-locked lasers," Optics Letters, vol. 32, pp. 626-628, Mar 2007. [15] R. Huber, D. C. Adler, and J. G. Fujimoto, "Buffered Fourier domain mode locking: unidirectional swept laser sources for optical coherence tomography imaging at 370,000 lines/s," Optics Letters, vol. 31, pp. 2975-2977, Oct 2006. [16] D. S. Adler, T. H. Ko, A. K. Konorev, D. S. Mamedov, V. V. Prokhorov, J. J. Fujimoto, and S. D. Yakubovich, "Broadband light source based on quantum-well superluminescent diodes for high- 183 resolution optical coherence tomography," Kvantovaya Elektronika, Moskva, vol. 34, pp. 915-8, 2004/10/ 2004. [17] D. C. Adler, T. H. Ko, and J. G. Fujimoto, "Speckle reduction in optical coherence tomography images by use of a spatially adaptive wavelet filter," Opt Lett, vol. 29, pp. 2878-80, Dec 15 2004. [18] D. C. Adler, T. H. Ko, P. R. Herz, and J. G. Fujimoto, "Optical coherence tomography contrast enhancement using spectroscopic analysis with spectral autocorrelation," Optics Express, vol. 12, pp. 5487-5501, Nov 1 2004. [19] T. H. Ko, D. C. Adler, J. G. Fujimoto, D. Mamedov, V. Prokhorov, V. Shidlovski, and S. Yakubovich, "Ultrahigh resolution optical coherence tomography imaging with a broadband superluminescent diode light source," Optics Express, vol. 12, pp. 2112-2119, MAY 17 2004. 184 Canadian Born We first saw light in Canada, the land beloved of God; We are the pulse of Canada, its marrow and its blood: And we, the men of Canada, can face the world and brag That we were born in Canada beneath the British flag. Few of us have the blood of kings, few are of courtly birth, But few are vagabonds or rogues of doubtful name and worth; And all have one credential that entitles us to bragThat we were born in Canada beneath the British flag. We've yet to make our money, we've yet to make our fame, But we have gold and glory in our clean colonial name; And every man's a millionaire if only he can brag That he was born in Canada beneath the British flag. No title and no coronet is half so proudly worn As that which we inherited as men Canadian born. We count no man so noble as the one who makes the brag That he was born in Canada beneath the British flag. The Dutch may have their Holland, the Spaniard have his Spain, The Yankee to the south of us must south of us remain; For not a man dare lift a hand against the men who brag That they were born in Canada beneath the British flag. -- Emily Pauline Johnson 1861-1913 185
