Echo Planar Spectroscopic Imaging and 31 P In Vivo Spectroscopy ECHO PLANAR SPECTROSCOPIC IMAGING AND 31 P IN VIVO MAGNETIC RESONANCE SPECTROSCOPY By SERGEI I. OBRUCHKOV, M.Sc. McMaster University ©Copyright by Sergei I. Obruchkov, March 2010 DOCTOR OF PHILOSOPHY (2010) McMaster University (Medical Physics and Applied Radiation Sciences) TITLE: Echo Planar Spectroscopic Imaging and Hamilton, Ontario 31 P In Vivo Magnetic Resonance Spectroscopy AUTHOR: Sergei I. Obruchkov, B.Sc.(Dalhousie University), M.Sc.(McMaster Uni- versity) SUPERVISOR: Professor M.D. Noseworthy NUMBER OF PAGES: xv, 178 ii Abstract The work in this thesis deals with pre–clinical development of rapid in vivo 31 P magnetic resonance spectroscopy (MRS) techniques. Current MRI literature of 31 P spectroscopy presents evidence of increased concentrations of phosphomonoesters (PME), and phosphodiester (PDE) as well as inorganic phosphate concentrations in tumor tissue. Human breast cancer studies have demonstrated correlation between disease progression and both PME and PDE peaks. Furthermore, 31 P MRS can be used to detect, grade tumours and monitor response to chemo and radiation therapy. Tumor measurements are typically static (i.e. single measurement per scan). In other experiments, on muscle for example, dynamic measures are required the purpose of which is to assess temporal function and recovery. In all 31 P acquisitions there are problems surrounding RF coil design, pulse sequence speed, localization and system calibration. The work presented here focused on improving all these aspects and provide easy and reliable work flow to use 31 P MRS in a clinical setting. One of the aspects of this thesis lies in designing and construction of an RF coil that is well suited for integration with a clinical MRI breast imaging and biopsy system. The designed coil was tuned for simultaneous operation at 31 P (51.73 MHz) and 1 H (127.88MHz) Larmor frequencies. This design has advantages in the fact that complex pulse sequences with heteronuclear decoupling could be performed easily. The additional features of the coil design is that it is possible to swap it into the breast imaging system without moving the patient. Along with the designed coil, custom software was written to assist with transmit gain calibration of 31 P RF pulses, to ensure maximum MR signal. The automated prescan ensures easy work flow and minimizes the operator variability and patient time inside the MR scanner. Another aspect of this thesis deals with rapid pulse sequence development, to further speed up the 31 P MRS data acquisition. Echo planar spectroscopic imaging (EPSI) with a fly–back gradient trajectory is currently one of the most reliable and robust techniques for speeding up chemical shift imaging (CSI) acquisitions. A 31 P EPSI sequence was written to acquire spectroscopic imaging data at 1, 2 and 2.6 cm spatial resolution and spectral bandwidth of 3125 Hz. The sequence showed an ability to speed up data acquisition up to 16 times, where SNR permits. Phantom studies were used to verify the double tuned coil and EPSI sequence ensuring proper and safe operation. In vivo measurements of an exercising muscle demonstrated the ability of 31 P EPSI to play an important role in rapidly acquiring spatially localized 31 P spectroscopic data. With these preclinical developments in place a clinical trial is possible using 31 P MRS rapidly and efficiently. Furthermore the increased usability of 31 P MRS provided by the tools developed in this thesis can prove to be beneficial by integrating 31 P MRS into existing clinical protocols. iii Acknowledgements I wish to thank my supervisor professor Michael D. Noseworthy, Ph.D for all his guidance, support and encouragement throughout this project. I also would like to thank my supervisory committee, Dr. Alex Bain, Dr. Charles Cunningham and Dr. Michael Patterson it is an honor to be guided by such bright people. There is an incredible number of people who helped me with this project. I would like to thank the entire Diagnostic Imaging Department of St. Joseph’s Healthcare. Thank you for taking me in as one of your own, I hope I gave back as much as I took. Toni Cormier thank you very much for all your help. A big thanks goes to Norm Konyer, who has shared his wisdom and allowed his calm personality to rub off on me. A very special thank you goes to Ken Bradshwaw from Sentinelle Medical who provided much needed direction in MR hardware development. Thanks to John Snider from GE Healthcare, for spending so much time with me and trusting me to take appart the scanner, you are an incredible resource that I will greatly miss. I would like to acknowledge the department of Medical Physics and Applied Radiation Sciences. I could not complete this thesis without their support. A special thanks goes out to the Dalhousie University Physics class of 2003, my development as a student and a physicist could not have had a better start. Thank you very much for still being there for me. And of course I would like to thank the most important people in my life, my parents Igor and Galina I couldn’t be here without your help. Thank you for all of your sacrifices and your unwavering support. My beautiful and incredible wife Heather I simply don’t know how I could do anything without you, thank you for your love, your support and your confidence in me. And lastly to my children Hali and Anya, thank you for enriching my life, you are my angels. iv Dedicated to the memory of Dr. Masayoshi Senba everyone can teach but not everyone can teach like you... v Table of Contents Abstract iii Acknowledgements iv List of Figures ix List of Tables xii List of Notations and Abbreviations xiii Chapter 1 1.1 1.2 1 General Overview And Introduction to in vivo Spectroscopy . . . . . 1 1.1.1 Clinical MRS . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1.2 Clinical MRS Techniques and Experimental Considerations . . 4 1.1.3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Multinuclear spectroscopy. . . . . . . . . . . . . . . . . . . . . . . . . 11 1.2.1 On the nature of spin . . . . . . . . . . . . . . . . . . . . . . . 11 1.2.2 Phosphorus NMR . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.2.3 Decoupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.2.4 Adiabatic RF . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.2.5 Echo Planar Spectroscopic Imaging (EPSI) . . . . . . . . . . . 20 1.2.6 The Problem and the Approach to a Solution . . . . . . . . . 22 Chapter 2 2.1 Introduction to In Vivo Spectroscopy 27 Theory Basic NMR Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.1.1 Quantum Description . . . . . . . . . . . . . . . . . . . . . . . 27 2.1.2 Classical Description . . . . . . . . . . . . . . . . . . . . . . . 30 2.1.3 Bulk Magnetization . . . . . . . . . . . . . . . . . . . . . . . . 31 2.1.4 NMR Signal Detection . . . . . . . . . . . . . . . . . . . . . . 33 2.1.5 Longitudinal and Transverse Relaxation . . . . . . . . . . . . 35 2.2 Chemical Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.3 J–Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2.4 Decoupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.5 Introduction to Imaging: Hard vs Soft RF excitation . . . . . . . . . 48 2.6 Adiabatic experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 53 2.6.1 64 Spatial Encoding . . . . . . . . . . . . . . . . . . . . . . . . . vi Chapter 3 jectory Echo Planar Spectroscopic Imaging of 31 P using Flyback Tra- 69 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.2 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 3.2.1 Chemical Shift Imaging signal equation . . . . . . . . . . . . . 71 3.2.2 Optimization of Echo–Planar Flyback Readout Trajectory Gradients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . 80 3.3 Methods and Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 3.4.1 Calibration Results . . . . . . . . . . . . . . . . . . . . . . . . 86 3.4.2 EPSI Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 3.2.3 3.5 Chapter 4 Imaging 4.1 Design and Construction of a Dual Tuned RF Coil for MR Breast Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Motivation for Dual Tuned Coil for Breast 31 P Spectroscopy With Two Port Input . . . . . . . . . . . . . . . . . . . . . . . 95 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.1.1 4.2 4.2.1 4.3 97 Resonating RLC Circuits . . . . . . . . . . . . . . . . . . . . . 101 4.2.3 Reciprocity Principle and Coil Quality Factor . . . . . . . . . 106 4.2.4 Transmit–Receive Switch . . . . . . . . . . . . . . . . . . . . . 108 4.2.5 Methods for Dual–Tuning Coils . . . . . . . . . . . . . . . . . 111 4.2.6 Dual Tuned Coil Design for Breast Imaging and Spectroscopy 116 Methods and Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 119 31 P Solenoid Coil For System Stability and QA Assurance Tests120 4.3.2 Dual Tuned Surface Coil Design and Construction . . . . . . . 123 4.3.3 B1 and B0 Field Calibration . . . . . . . . . . . . . . . . . . . 125 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 4.4.1 4.5 Propagation of Electromagnetic Waves and Magnetic Field Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 4.3.1 4.4 93 Coil Performance Imaging, Spectroscopy and Decoupling . . . 128 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 vii Chapter 5 Design and Construction of an inexpensive Polyvinyl Alcohol Breast Like Phantom for use with MR Imaging 135 5.1 Introducion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 5.2 Methods and Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 137 5.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 140 Chapter 6 Dynamic Measurements of 31 P Metabolites in a Posterior Compartment of the Leg During Plantar Flexion Exercise Using Surface Coil or Echo Planar Spectroscopic Imaging for Localization 143 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 6.2 Methods and Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 146 6.3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 149 Chapter 7 Conclusions and Future Directions 155 7.1 Future Applications of the Constructed Double Tuned 1 H – 31 P Breast RF Coil and the Designed flyback 31 P EPSI Sequence . . . . . . . . . 155 7.2 Future Directions and other Applications of ESPI sequence and Multi Tuned MR Coils. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Bibliography 161 Appendix A 169 Appendix B 175 Appendix C 177 viii List of Figures 1.1 RF Coils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.2 Nuclear structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3 31P Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.4 EPSI PSD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.1 Zeeman Splitting of Energy Levels . . . . . . . . . . . . . . . . . . . . . 29 2.2 Spin Precession of 1 H Nuclei in Magnetic Field Bo . . . . . . . . . . . . 30 2.3 Contribution of individual 1 H Nuclei Magnetic Moments to Bulk Magnetization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.4 Magnetization in Lab vs Rotating Frames of Refference . . . . . . . . . . 34 2.5 Longitudinal and Transverse Relaxation . . . . . . . . . . . . . . . . . . 39 2.6 Spin–Echo Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.7 Chemical Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2.8 J–Coupling between two heteronuclear systems energy levels. . . . . . . . 43 2.9 J–coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.10 Slice Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.11 RF Coils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 2.13 NMR a reversible adiabatic experiment . . . . . . . . . . . . . . . . . . . 56 2.14 Effective Field created by the Ofresonance effects . . . . . . . . . . . . . 61 2.15 Adiabatic Pulses and Their Profile . . . . . . . . . . . . . . . . . . . . . 63 2.16 Spatial Encoding of 2D Frequency Space and Fourier Transformation . . 67 3.1 2D CSI Pulse Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.2 2D CSI data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.3 Comparing Point Spread Functions . . . . . . . . . . . . . . . . . . . . . 75 3.4 EPSI Pulse Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.5 Segmented Flyback EPSI sequence . . . . . . . . . . . . . . . . . . . . . 77 3.6 Phase offset of the EPSI acquisition . . . . . . . . . . . . . . . . . . . . . 79 3.7 3.0 Tesla MRI Scanner . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 3.8 Linear Readout with Dead Time . . . . . . . . . . . . . . . . . . . . . . 84 3.9 Synchronising Gradients and Data Sampling . . . . . . . . . . . . . . . . 85 3.10 Interleaving Gradient Waveforms . . . . . . . . . . . . . . . . . . . . . . 86 ix 3.12 Overlay of Echoes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 3.13 CSI and EPSI Data of a Spherical Phantom . . . . . . . . . . . . . . . . 89 3.14 High Resolution CSI and EPSI Data . . . . . . . . . . . . . . . . . . . . 90 3.15 SNR Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.1 Vanguard Breast Biopsy Table . . . . . . . . . . . . . . . . . . . . . . . . 96 4.2 Field generated from a circular loop . . . . . . . . . . . . . . . . . . . . . 98 4.3 Field Profile of a Surface Coil . . . . . . . . . . . . . . . . . . . . . . . . 100 4.4 Resonating RLC Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 4.5 Tuning and Matching Capacitors in MR coil . . . . . . . . . . . . . . . . 106 4.6 Reactance and Resistance of the RF Coil . . . . . . . . . . . . . . . . . . 108 4.7 Linear and Hybrid TR switch schematics . . . . . . . . . . . . . . . . . . 110 4.8 Circuits for dual tune approaches . . . . . . . . . . . . . . . . . . . . . . 111 4.9 Two Coil Approach to Dual Tuning . . . . . . . . . . . . . . . . . . . . . 113 4.10 Crossed Elements Dual–Tuned Coils . . . . . . . . . . . . . . . . . . . . 115 4.11 Layout of Orthogonal Dual Tuned Birdcage Coils . . . . . . . . . . . . . 115 4.12 Circuit for a Dual Tuned Coil . . . . . . . . . . . . . . . . . . . . . . . . 116 4.13 Reactances of RLC, Tank and dual tuned circuits . . . . . . . . . . . . . 117 4.14 Decoupler Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 4.15 RF Coil Dock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 4.16 Simulated Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 4.17 Physical Layout of the Coil . . . . . . . . . . . . . . . . . . . . . . . . . 124 4.18 Varying value of L2 inductor . . . . . . . . . . . . . . . . . . . . . . . . . 125 4.19 Dual Tuned Coil and Sentinelle Breast Biopsy Table . . . . . . . . . . . 125 4.20 TG Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 4.21 High Order Shimming data . . . . . . . . . . . . . . . . . . . . . . . . . 127 4.22 Bench Testing Results of a Dual Tuned Coil . . . . . . . . . . . . . . . . 129 4.23 B1 Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 4.24 Gradient Echo Excitation Profile . . . . . . . . . . . . . . . . . . . . . . 130 4.25 Spin Echo Excitation Profile . . . . . . . . . . . . . . . . . . . . . . . . . 131 4.26 Decoupling of Methylphosphonic Acid . . . . . . . . . . . . . . . . . . . 132 5.1 Polyvinyl alcohol molecule . . . . . . . . . . . . . . . . . . . . . . . . . . 137 x 5.3 T1 and T2 Weighted Phantom MR Images . . . . . . . . . . . . . . . . . 141 6.1 Schematic of the biochemical reaction in muscle bioenergetics . . . . . . 146 6.2 CSI and EPSI Data Overlaid Over 1 H Image of a Leg . . . . . . . . . . . 149 6.3 Dynamic Study of Plantar Flexion Exercise Amplitude Change . . . . . . 150 6.4 PCr and Pi in an Exercising Leg . . . . . . . . . . . . . . . . . . . . . . 151 6.5 Dynamic Study of Plantar Flexion Exercise Spectral Frequency Change . 152 6.6 pH Calculation of 31 P MRS in an Exercising Leg Without Spatial Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 xi List of Tables 1.1 Phosphorus chemical shift . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3.1 In vivo concentration of phosphorous metabolites in human brain tissue. 71 3.2 SNR is proportional to square root of total acquisition time . . . . . . . 73 3.3 1.5 T and 3.0 T Magnet Parameters . . . . . . . . . . . . . . . . . . . . 82 3.4 List of Gradient Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 5.1 T1 and T2 of 10% and 15% PVA cryogel . . . . . . . . . . . . . . . . . . 141 6.1 Relaxation Times of 31 P Metabolites and Their Chemical Shifts . . . . . 146 xii List of Notations and Abbreviations R1 Longitudinal relaxation constant of a contrast agent T1 Longitudinal relaxation time T2 Transverse relaxation time χ Magnetic susceptibility CEMRA Contrast enhanced magnetic resonance angiography CNR Contrast to noise ratio CT Computed Tomography DC Direct current ETL Echo train length FID Free induction decay FOV Field of view FSE Fast spin echo GE General Electric GRE Gradient recalled echo Gd-DTPA-BMA Gadolinium complex of diethylenetriaminepentaacetic acid bismethylamide. Gd Gadolinium IR Inversion recovery LV Levenberg-Marquardt MIP Maximum intensity projection MRA Magnetic resonance angiography MRI Magnetic resonance imaging MTT Mean transit time MVA Minimal variance algorithm xiii NEX Number of excitations NMR Nuclear magnetic resonance PE Phase encoding PR Projection reconstruction RF Radio Frequency ROI Region of interest SE Spin echo SNR Signal to noise ratio HOS High Order Shimming SPGR Spoiled gradient recalled SoS Sum of squares TE Time to echo TR Repetition time TTP Time to peak T Tesla bpm Beats per minute cc Cubic centimeter ml Milliliter mM Millimolar mm Millimeter ms Milliseconds s Second xiv E Electric Field Intensity (volts/meter) D Electric Flux Density (coulombs/meter2 ) H Magnetic Field Intensity (amperes/meter) B Magnetic Flux Density (webers/meter2 ) J Electric Current Density (amperes/meter2 ) ρ Electric Charge Density (coulombs/meter3 ) 0 Free space permitivity (farads/meter) µ0 Free Space Permeability (henrys/meter) σ Conductivity (siemens/meter) xv xvi PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 1 Introduction to In Vivo Spectroscopy 1.1 General Overview And Introduction to in vivo Spectroscopy Nuclear magnetic resonance theory describes the physical interaction of particles with a nuclear magnetic dipole moment µ̂ in a magnetic field Bo . In vivo NMR refers to a technique which enables identification and quantification of a large number of biologically important molecules in the intact tissue of a living organism (some good references to in vivo NMR are de Certaines et al. (1992), de Graaf (2005a) and de Graaf (1998)). In vivo NMR also known as magnetic resonance spectroscopy (MRS) is used as a diagnostic tool on a daily basis world wide. The first in vivo spectra were obtained from rat brain in 1983 by Behar at Yale University, using a modified NMR apparatus. Soon after, papers by Lauterbur and Mansfield appeared describing in vivo NMR imaging concepts. Applying position dependent magnetic field gradients in addition to the static field allowed Lauterbur and Mansfield to spatially localize NMR signal and construct an image. Soon after the first successful experiments, apparatus for clinical applications became available. The first hardware for producing images was big and bulky, and imaging was slow as only a few pulse sequences were available. First clinical diagnosis were done largely by measuring T1 1 PhD Thesis Sergei Obruchkov Dept. of Medical Physics and T2 times of tissues. Today in vivo NMR got renamed as simply MRI, allowing images to be acquired with high soft tissue contrast and in vivo spectroscopy can be performed from any particular volume within the tissue using accurate localization techniques. Clinical diagnosis today can be performed by much more advanced techniques than was first thought possible: for example, assessment of diffusion of water inside the tissues using diffusion tensor and diffusion weighted imaging (DTI and DWI, respectfully), functional brain imaging using fMRI, angiography using contrast agents, susceptibility weighted imaging (SWI) allowing for venous mapping and observing calcium and iron deposits, and spectral editing techniques for studying metabolic concentrations of lactate and gamma-aminobutyric acid (GABA) in vivo, just to name a few. The following sections will describe in more detail clinically available magnetic resonance spectroscopy (MRS) techniques, current approaches and challenges. 1.1.1 Clinical MRS Proton magnetic resonance spectroscopy or hydrogen spectroscopy (1 H-MRS) is the most clinically applied spectroscopy technique today. The proton is one of the most NMR sensitive nuclei not only in terms of high gyromagnetic ratio but also of its natural abundance (99.9%) and concentration (i.e. since virtually all metabolites contain protons). Proton spectroscopy has been most successfully applied in identifying and quantifying cerebral metabolites and is considered a valuable diagnostic tool in diag2 PhD Thesis Sergei Obruchkov Dept. of Medical Physics nosing and grading many diseases including metabolic disorders, and cancer. Even though the sensitivity of in vivo 1 H NMR is limited to the detection of metabolites with concentrations above 0.5 mM (de Graaf 2005a) it has still proven invaluable in diagnosing aberrant metabolic processes. The success of proton spectroscopy, in many parts of the body, has been limited due to technical difficulties and limited information content e.g. liver and muscle spectra are relatively simple in comparison to brain spectra due to large dominating lipid resonances. In addition, liver spectroscopy is challenging due to high iron concentrations and respiratory motion which both contribute to broadening of resonance lines. Muscle 1 H spectroscopy is of interest as there are significant differences in metabolic content between muscle groups and metabolic states (e.g. rest/exercise). In addition, metabolite resonances such as carnosine 2 and 4 exhibit chemical shift dependence on pH enabling the use of 1 H spectroscopy as an intracellular pH meter (Moon and Richards 1973; Pan et al. 1988). With the introduction of commercially available pulse sequences optimised for prostate and breast cancer studies 1 H MRS is undoubtedly a growingly useful technique that can be applied to all parts of the body. Other NMR visible nuclei (see section 1.2.1) such as 31 P, 23 Na, 13 C, 7 Li, 19 F and 39 K are either used rarely or almost never used for clinical spectroscopy. There are many reasons for this, the first being that most commercial clinical MR systems are designed to work only with 1 H nuclei (i.e. what is only required for clinical imaging). Hardware for “multi” nuclear work is available as an extension to most existing clinical systems 3 PhD Thesis Sergei Obruchkov Dept. of Medical Physics however rough estimate puts less then 5% of all MRI systems in Canada capable of this functionality∗ . Section 1.2 will deal with multinuclear spectroscopy in particular, for now I will focus on MRS in general bringing examples from the clinical application of proton spectroscopy. 1.1.2 Clinical MRS Techniques and Experimental Considerations Obtaining MR images and MR spectra depend on selective excitation of a certain volume from which signal is recorded. This is accomplished by manipulating magnetic gradients and RF pulses in order to excite a particular volume of interest. The signal from the excited volume is acquired, digitised and processed via Fourier transform. Following acquisition, data reconstruction results get processed via a semi-automatic analysis and the end result is diplayed. State of the art MR systems today routinely present only a simple analysis, and are often incapable of automatically correcting for errors if the spectra are deemed of poor quality. From the clinical perspective the technical process is transparent to the end user. Clinical users are given only an ability to choose the volume of interest on the anatomy and specify the most basic NMR related parameters. At first sight clinical MRS seems straight forward. In reality, however, clinical MRS acquisition is incredibly complex. Even if all the technical problems could be solved, challenges of patient positioning, patient cooperation, voxel placement and diagnosis are still great. Thus, even though ∗ Currently there are approximate;y 200 MRI systems in Canada, therefore around 10 have MNS capabilities. 4 PhD Thesis Sergei Obruchkov Dept. of Medical Physics the data from MRS is so rich it is simply not as widely clinically accepted as one would imagine. There are a few reasons as to why wide use of clinical spectroscopy is limited. First, the acquisition spectral artifacts can be hard to identify and often require a physical scientist specialising in spectroscopy to perform analysis. Secondly achieving absolute metabolite concentrations (and how these can be related to clinical conditions) is not established, or even completely agreed upon, which results in many between-study, and even between-patient variations. Spectral Localization There are two types of single voxel spectral localisation approaches: multi shot techniques, such as image selected in vivo spectroscopy (ISIS) and single shot techniques such as stimulated echo acquisition mode (STEAM) or point resolved spectroscopy (PRESS). In addition, an array of other techniques have previously been used, for example volume selective spectroscopy (VOSY) and volume restriction using RF surface coils. In current clinical practice however, STEAM and PRESS are considered the “tools of the trade”. Both techniques utilise three slice selective RF pulses and gradient combinations exciting three orthogonal slabs in such a way as to generate an echo (MR signal) from an intersecting volume. The PRESS technique utilises a combination of 90o -180o -180o (slice selective pulses) to produce a spin-echo while the STEAM sequence use a 90o -90o -90o combination to 5 PhD Thesis Sergei Obruchkov Dept. of Medical Physics produce a stimulated echo. Over a single TR the PRESS pulse sequence is usually longer (i.e. longer TE) than STEAM and thus has difficulty visualizing metabolites with short relaxation times. The PRESS sequence is known to better visualize metabolites with longer T2 relaxation times, is less susceptible to motion, diffusion, and quantum effects, and has a theoretically better SNR than STEAM over the same voxel. All of the above described techniques are known as single voxel spectroscopy as they are used to acquire data from a single volume of interest (VOI). For more information on localisation techniques please see the paper by Keevil (2006). A comparison between STEAM and PRESS, in terms of variability in metabolite measurements can be seen in Kim et al. (2005). Although a single voxel acquisition is tremendously useful for understanding in vivo metabolic processes, often comparison between diseased and healthy tissues, or a study of regional metabolite variation is needed. With single voxel scan times often approaching 10 minutes it is not clinically feasible to perform multiple single voxel acquisitions. Hence, multi-voxel, or chemical shift imaging (CSI) was developed. Chemical shift imaging is an extension of the single voxel MRS technique used to obtain data from multiple voxels. This may be achieved essentially by sequentially acquiring data from adjacent voxels to produce spectral maps in 2D and 3D. Details of the CSI approach is discussed further in section 1.2.5. 6 PhD Thesis Sergei Obruchkov Dept. of Medical Physics RF Coils Radiofrequency (RF) coils are used to obtain MR signals. These are high Q resonating circuits tuned to a specific resonating frequency, dependent on the nucleus (more specifically the nuclear gyromagnetic ratio), and the strength of magnetic field. There are two types of coils: surface coils and volume coils. Volume coils are typically large assemblies in a Helmholtz, solenoid, or birdcage geometry, that can produce a uniform B1 RF field which is used to excite the spins. Surface coils are much more efficient at receiving a signal close to the coil but their effectiveness significantly decreases at about half the coil radius away. They are rarely used for excitation as they do not produce a uniform RF B1 field (see Figure 2.11). Therefore current approaches in MR utilize an array of receive-only surface coils that can be positioned close to the patient’s body (i.e. ensuring good coupling) to achieve a large volume coverage to receive MR signal. Volume coils are used for excitation purposes. Volume coils and arrays are superior for imaging applications. However, the use of coil arrays for spectroscopy adds a layer of complexity as it becomes non-trivial to add multiple broadband spectroscopic signals together. Specifically, phase correction and scaling factors during the adding process may affect quantification during the data analysis. Relaxation, Diffusion, Coupling Effects and Other Distortions. MR spectra are affected by user parameter selections: repetition time (TR), echo time (TE), location of the voxel within the anatomy, and other factors such as mag7 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 1.1: B1 field plots of a birdcage coil (right) and surface coil cross–section (left). The flip angle iso-lines clearly show a region of uniform excitation that can be achieved using bird cage geometry. netic field uniformity (∆B0 , ∆B1 ), patient motion, and temperature to name a few. Nevertheless MR spectroscopy is clinically extremely useful, as long as the effects and artifacts are understood and identified during data analysis to ensure correct diagnosis. TR and TE times set by the user affect spectral appearance as different species relax with different T1 and T2 relaxation times. Ideally a very short TE and a very long TR, are best for capturing all of the metabolites on a spectrum. However this is often disadvantageous as the spectrum would be extremely crowded and hard to interpret plus patient scan time would be unreasonable long. Often longer TE times (e.g. 144, 288 ms) are used in order to let species with shorter T2 times relax and produce spectra with only certain metabolites. Molecular diffusion during the delays between RF pulses and application of magnetic field gradients results in signal contamination from spins that are not properly “tagged”. This produces a spectrum which is difficult to quantify, especially when comparing molecules with large diffusion coefficients. Minimizing TE, using smaller 8 PhD Thesis Sergei Obruchkov Dept. of Medical Physics gradients, or using a pulse sequence that is insensitive to diffusion e.g. utilising Carr-Purcell-Meiboom-Gill (CPMG) echoes instead of Hahn echoes can be used to eliminate this effect (de Graaf 2005a). Other factors that affect spectroscopy include scalar coupling between different nuclei e.g. 1 H–13 C, and the chosen water/fat suppression techniques used. To add more to the complexities, angular orientation of structured tissue in the magnetic field, truncated sampling causing “voxel bleeding”, blood pulsatility near the voxel and many more factors not mentioned here tend to complicate the life of an in vivo NMR scientist. 1.1.3 Data Analysis Fourier transformed NMR, in principle, can be used as a quantitative technique to derive absolute metabolite concentrations in tissues. The area under the curve of the FFT data is related to the longitudinal thermal equilibrium magnetization vector, which in turn is proportional to the number of spins in the sample. True quantification of the data however is challenging as often the RF field (B1 ) varies spatially throughout the sample. In addition, in vivo NMR lacks a true internal or external reference for quantification: internally all metabolites can potentially vary, while external references (e.g. vials of chemical solutions placed on the subject’s skin) do not experience the same B1 and B0 fields as the voxel of interest. Proton spectroscopy often utilises concentration of the endogenous metabolite creatine in the brain. The 9 PhD Thesis Sergei Obruchkov Dept. of Medical Physics reported concentrations of other metabolites are then reported relative to the creatine peak. But absolute concentration cannot truly be calculated this way due to large variability of creatine within people, and also between ages and diseases. The external reference technique utilises a small vial with a known concentration of metabolites. Thus this approach requires acquisition of at least two spectra; one from the VOI another from the reference sample. The RF inhomogeneity between the two locations can be accounted for if the coil’s RF field distribution is known. The analysis of spectra often requires input from the user, as automated techniques either do a poor job (i.e. do not work on all spectra due to poor SNR, artifacts, mysteriously varying baselines, etc.) or require previous knowledge such as a database of standards. There are a few approaches to automated spectral analysis, the most successful of these techniques utilising a least squares fitting algorithms. The general principle of these methods is that the fitted “theoretical” spectrum should resemble the experimental spectrum as closely as possible by varying parameters Mo , T2 , ∆ω, φo for each resonance. The two most frequently used algorithms are VARPO (Vanhamme et al. 2001), (Mierisova and Ala-Korpela 2001) a time domain algorithm, and LCmodel (Provencher 1993), a frequency domain algorithm. The time domain method is flexible enough to quantify frequency shifts and can avoid problems with intense baselines, while frequency domain analysis is most often utilised for more complex spectra, such that are observed with short echo times. However, the LCmodel technique can perform analysis only for specific echo times TE=30, 35, 50 and 144ms. Although if other echo times are desired new basis sets can be devel10 PhD Thesis Sergei Obruchkov Dept. of Medical Physics oped (a time consuming process). Automated spectral analysis approaches, without a priori knowledge are quite poor as results have to fit within 95% of a certainty calculation otherwise the convergence will fail. Further, to avoid baseline issues, some methods, instead of integrating peaks, use convolution and deconvolution, to collapse the peaks into a series of delta like functions the heights of which are proportional to peak “areas under the curve”. This approach reports relative heights of the peaks as the relative concentrations. 1.2 Multinuclear spectroscopy. Clinical implementation of Nuclear Magnetic Resonance (NMR) is largely focused on 1 H nuclei mostly due to their abundance and biological importance. There is a vast number of other NMR visible nuclei. The visibility of a nuclear isotope is dependent on its atomic structure. This section will focus on 31 P in vivo spectroscopy. 1.2.1 On the nature of spin Spin I is an intrinsic property of a fundamental particle. In order to understand why some atoms have spin and other do not, a brief review of subatomic particles and their properties is necessary (see Figure 1.2). The classical view of an atom consists of a positively charged nucleus surrounded by a cloud of orbiting negatively charged electrons. The nucleus of an atom always has a positive net charge, and is made up from neutrons and protons. However the neutron and proton are not fundamental 11 PhD Thesis Sergei Obruchkov Dept. of Medical Physics atomic particles, as if either is broken apart smaller subatomic particles, called quarks, will be released. Quarks come in six different flavors and three colors. Both neutron and proton each consist of three quarks that are held together by gluons. From Figure 1.2 it is noted that the neutron consists of two down quarks and one up quark, adding up the charges of these three quarks −1/3e − 1/3e + 2/3e = 0 gives a neutron net zero charge, while adding the charges of three quarks that make up a proton we end up with a net charge of −1/3e + 2/3e + 2/3e = +1e, where e is electron charge. Performing additions with quark spins results in a net spin of the proton and neutron calculated as I = 21 , however a proton has a positive spin and neutron negative spin. Since most nuclei are made up from multiple neutrons and protons, it is possible to divide them into three categories: • Even-even nuclei: Having an even number of protons and neutrons, the spins cancel out producing a net spin of zero. • Even-odd and odd-even nuclei: The final spin of nuclei with uneven numbers of protons and neutron is determined by the single unpaired nucleon. • Odd-odd: This type of nucleus is harder to describe as the unpaired protons and neutrons produce a net spin that is determined by the parallel and anti-parallel alignment of the proton and neutron spins. Nuclei with spin have net positive or net negative spin depending on the excess of protons or neutrons. The sign of the spin has only one effect on a nucleus in a magnetic field: positive spins will align with the magnetic field and will precess in 12 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 1.2: Modern view of a nucleus. a clockwise direction, while a nucleus with negative spin will align anti-parallel with the magnetic field and precess counter clockwise around the magnetic field. 1.2.2 Phosphorus NMR The success of proton NMR in clinical applications has only been matched by phosphorus 31 P NMR. A high natural abundance and sensitivity of around 7% that of proton, allows acquisition of a good quality spectrum within clinically relevant/acceptable time scales (i.e. minutes). A large chemical shift dispersion of 30 ppm, and considerably fewer metabolites than 1 H-NMR, produces “good” peak resolution, see table 1.1. Most importantly 31 P spectroscopy is capable of detecting all of the 13 PhD Thesis Sergei Obruchkov Dept. of Medical Physics key compounds involved in energy metabolism. Furthermore the chemical shift of some phosphorus containing compounds is sensitive to the concentrations of magnesium and pH. This property can be explained by the chemical exchange phenomenon, where during the protonation with magnesium, changes in the local magnetic environment of 31 P nucleus occur. Chemical exchange creates two species of the compound e.g. protonated and unprotonated, with two distinct chemical shifts but because this exchange process is rapid in contrast to the NMR time scale only an average chemical shift of the two species is observed. Clinically relevant pH measurements are performed on the inorganic phosphate peak using a Henderson-Hasselbach relationship: pH = pKA + log δ − δHA δA − δ (1.1) were δ is the observed chemical shift, δHA and δA are protonated and unprotonated chemical shifts, and pKA is a equilibrium constant of a compound, (pK=6.77 for inorganic phosphate) (Hoult et al. 1974; Moon and Richards 1973; Pan et al. 1988). The calculations of free magnesium concentrations is often deduced similarly by using equation 1.1 from the chemical shift of the β-ATP dominant peak. Phosphorus spectroscopy has been also applied for non invasive measurements of the creatine kinase enzymatic reaction (de Graaf 2005a). P Cr2− + M gADP − + H + ←→ M gAT P 2− + Cr 14 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Table 1.1: Phosphorus chemical shifts for low molecular weight metabolites, at pH 7.2 and fully complexed with magnesium (Shulman et. al. 2005). Typically spectral width on a clinical 1.5T spectrometer of 500Hz is enough to acquire all of the 1 H resonances, phosphorus on the other hand requires about 3 kHz. 31 Compound Adenosine monophospate (AMP) Adenosine diphosphate (ADP) α β α β γ Adenosine triphospate (ATP) Dihydroxyacetone phosphate Fructose-6-phosphate Glucose-1-phosphate Glucose-6-phosphate Glycerol-1-phospate Glycerol-3-phosphorycholine Glycerol-3-phosphorylethanolamino Inorganic Phosphate Nicotinamide adenine dinucleotide (NADH) Phosphocreatine Phosphoenolpyruvate Phosphorylcholine 31 P chemical shift (ppm) 6.33 -7.05 -3.09 -7.52 -16.26 -2.48 7.56 6.64 5.15 7.20 7.02 2.76 3.20 5.02 -8.30 0.00 2.06 5.88 P NMR spectra are comparatively simple (relative to proton spectra) with reso- nances from phosphocreatine (PCr), adenosine triphosphate (ATP), inorganic phosphate (Pi) and phosphomono and phosphodiesters. In vivo 31 P NMR can be used to probe dynamic changes of rapid and slow high energy phosphate reactions. For example, the chemical shift of the Pi and the β–ATP resonances are dependent on the pH and magnesium concentration. These exchange reactions are typically very fast as they are not catalyzed by enzymes. Many of the other compounds that are typically observed by 31 P NMR are very important in enzyme–catalyzed reactions, such as: ATP synthesis and the interconversion of PCr and ATP by creatine kinase. Combin15 PhD Thesis ing 31 Sergei Obruchkov Dept. of Medical Physics P MRS with magnetization transfer (MT) experiments have been extensively used to quantitatively study metabolic pathways (Forsen and Hoffman 1963; de Graaf 2007; Brown et al. 1977). An extensive amount of literature exists describing how 31 P MRS has been used to assess phosphorous containing metabolites in healthy and diseased brain, heart, liver and muscle tissues (see section 6). Adult brain is highly (and exclusively) aerobic, which results in high rates of oxygen consumption. Glucose is the fuel of choice, in the healthy brain. Glucose is metabolized via glycolysis and subsequent tricarboxylic acid cycle to form a maximal amount of ATP. The resultant energy is stored in the high energy phosphate bonds of ATP. The energy is used (i.e. hydrolysis of high energy phosphate bonds) to maintain ongoing electrophysiological activities in the resting brain, and to supply additional energy for increased neuronal activity during brain activation and task performance. 31 P MRS offers a non–invasive neuroimaging technique that enables quantitative as- sessment of the cerebral metabolic rates of ATP in both resting and activated brain states. MR imaging and spectroscopy has an important role in helping understand muscle pathologies. More importantly, 31 P MRS has been used to assist in drug development and evaluation. Recent research in development of gene therapy treatments are being introduced (such as gene and stem cell therapies) for which MR methods may be very useful to monitor the effect. MR biomarkers have been proposed that can provide feedback information on the treatment, one of which uses arginine kinase as a reporter 16 PhD Thesis gene and the Sergei Obruchkov 31 Dept. of Medical Physics P signal of phospho-arginine as biomarker (Walter et al. 2000; Leroy- Willig et al. 2003). Several 31 P MRS detectable metabolic changes have been observed following anti- cancer treatment. Both radiotherapy and DNA-damaging chemotherapy have shown elevation in the phospholipid precursor phosphocholine (PC) in tumors compared to normal tissue. Furthermore, this lowers following treatment, frequently prior to tumor shrinkage (Gillies and Morse 2005). This observation can be explained in part by inhibition of cellular proliferation (Ronen et al. 1992). 31 P MRS is also frequently used to assess tumor energy metabolism reporting on parameters such as the levels of ATP, inorganic phosphate and sugar phosphate intermediates of glycolysis, and tissue pH. Typically, a large reduction in ATP and intracellular pH, as well as increased levels of inorganic phosphate, are observed with successful drug therapies (Beloueche-Babari et al. 2010). Therefore 13 31 P and MRS of other nuclei such as 1 H, C and 17 O are highly valuable as noninvasive longitudinal biomarkers that can pro- vide information on success of drug delivery and efficacy of drug-target treatments. These unquestionably provide unique non invasive feedback that aid in therapeutic guidance and allow rapid identification of resistance to therapies. 1.2.3 Decoupling Coupled nuclear spins exchange energy in the same manner as two pendulums attached by a spring do: the spring causes them to swing intermittently. In NMR this 17 PhD Thesis Sergei Obruchkov Dept. of Medical Physics effect causes spectral splitting. This often causes an additional complexity for in vivo NMR as the peaks become overlapped and quantification becomes virtually impossible. This is classically seen as a problem in in vivo 1 H-NMR . For 31 P-NMR there is clear separation in resonances observed in muscle while in brain there are overlapping resonances, particularly in the phosphate mono and diesters (figure 1.3). The process of removing the splitting between spins is called decoupling. Decoupling is achieved with the aid of a RF selective saturation pulse. Often compound pulses i.e. a set of 90, 180, 270 pulses with varied phase cycling approaches are used to saturate hydrogen spins, resulting in collapse of the splitting, thus eliminating the added complexity. 1.2.4 Adiabatic RF In vivo NMR uses radio frequency pulses that are used simultaneously for localisation and excitation. When surface coils are used for excitation it becomes difficult to excite a sample uniformly due to inhomogeneous distribution of the B1 RF field. Adiabatic RF pulses however achieve uniform excitation by sweeping the amplitude of the pulse sinusoidally while sweeping the frequency as a cosine function. Two types of adiabatic pulses exist; selective and non-selective. Non-selective pulses are the easiest to design and implement, while selective adiabatic pulses are much more sophisticated and require one to simultaneously modulate the RF phase and amplitude, and the amplitude of the magnetic field gradients. 18 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 1.3: 31 P spectra from different parts of the body. Adenosine triphosphate (ATP) is the energy currency in living systems. Phosphocreatine (PCr), another high energy phosphate, is used for storing energy and converting ADP to ATP. It is absent in liver, kidney and mammalian red blood cells. Inorganic Phosphate (Pi ) is generated from hydrolysis of ATP and often seen increased in compromised tissues. Phosphomonoester (PME) signal contains contribution from membrane phospholipid constituents, glucose-6-phosphate and glycerol-3 phosphate. Phosphodiester (PDE) signal contains contribution from membrane constituents. A good review of adiabatic full passage pulses for spectroscopic imaging has been written by Sacolick et al. 2007. There are many different variations of selective excitation from Carr-Purcell adiabatic excitation, localization by adiabatic selective refocusing (LASER) (Garwood and DelaBarre 2001), B1 insensitive universal rotator BIR-4 (Staewen et al. 1990), B1 insensitive slice selection using 8 adiabatic segments BISS-8 (de Graaf et al. 1996), and other varieties each with their own advantages and disadvantages. 19 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 1.2.5 Echo Planar Spectroscopic Imaging (EPSI) Chemical shift imaging (CSI), is a process of acquiring 2D or 3D adjacent multiple voxels, often using standard PRESS or STEAM localization. The general idea is to use a train of 3 slice selective RF pulses to localise a volume, during which time spectral information is lost due to T2 and J-coupling effects. It takes a long time to obtain multiple voxels using phase encoding acquisition of individual voxels using PRESS CSI or STEAM CSI pulse sequences. In order to speed up the process faster techniques can be used. Echoplanar spectroscopic imaging (EPSI) was originally proposed by Mansfield in 1984 motivated by the need for an in vivo spectroscopy method with a short TE time. Posse et al. (1994) successfully demonstrated acceleration of spectral acquisition of a 3D data set using EPSI. EPSI unlike PRESS/STEAM CSI utilises only one slice selective RF pulse, and the localization in the spatial dimension is performed by a sweeping magnetic gradient i.e. kx see figure 1.4. Gradient reversal during the echo planar readout creates “odd” and “even” echoes, as the data is sampled during the forward and during the fly-back readout. In theory, odd and even echoes should be equivalent after a phase reversal and correction. However, technical limitations make odd and even echoes incompatible with each other often due to timing errors, low spectral bandwidth, eddy currents, susceptibility and other artifacts (Chen et al. 2007; Cunningham et al. 2005). Often spectral spatial oversampling is performed in order to compensate, with an added requirement 20 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 1.4: Basic Design of the EPSI pulse sequence, slice selection and refocusing can be in any form: e.g. gradient echo, Hahn echo, or CPMG echo, each with their own advantages and disadvantages. Variations of EPSI utilise additional 1 H nuclear Overhauser effect (NOE) pre-pulses to increase SNR. Furthremore, decoupling for 31 P and 13 C are often performed as well. of extra encoding steps during the EPI readout. Therefore fly-back techniques are considered more reliable and do not increase total scan time. EPSI essentially accelerates the acquisition of spectral and spatial information. Consider a scan requiring 32x32 spatial points and 512 spectral points, EPSI will acquire all of the data in 32 TR steps. In comparison, conventional CSI techniques used clinically will acquire the same amount of data in 32x32 TR steps, or 32 times longer. For excitation purposes a single channel volume coil is typically used, while the acquisition is performed using an array of small surface coils. Depending on geometric distribution of the surface coils parallel acquisition techniques such as sensitivity encoding (SENSE) and generalized autocalibrating partially parallel acquisitions (GRAPPA) can be used for accelerating spatial encoding. These techniques have 21 PhD Thesis Sergei Obruchkov Dept. of Medical Physics been significantly explored for accelerating MR imaging. 1 H and multinuclear spectroscopy can benefit from this speed-up, and has recently been demonstrated with proton spectroscopy (Lin et al. 2007; Zhu et al. 2007). Currently there is very little acceleration with SENSE or GRAPPA done with in vivo multinuclear NMR. 1.2.6 The Problem and the Approach to a Solution Magnetic resonance spectroscopy is a valuable method for evaluation of tissue metabolism. Current clinical spectroscopy techniques acquire static snapshots of metabolite concentrations in tissues. This “static” data is limited in information and is not very representative of changes that occur on a physiological time scale (i.e. less then a second). While some single voxel spectroscopy can be performed as fast as within 10-20 seconds, current multivoxel acquisitions are not similarly as fast. Thus single voxel placement is relied upon, which demands accurate voxel placement (i.e. to properly sample the diseased area). This requires previous knowledge about the diseased location, which would necessarily be obtained [possibly] from other imaging modalities. Therefore a multivoxel approach is more desirable as it would allow simultaneous comparison between healthy and diseased tissue regions, and it would display localisation and differentiation of the diseased tissue. While 1 H is an important nucleus the project will focus on 31 P as it is extremely useful to observe dynamic changes in high energy metabolism and also intracellular pH. 22 PhD Thesis Sergei Obruchkov Dept. of Medical Physics EPSI is a fast spectroscopic technique which allows acquisition of localised in vivo MR spectra, and is ideal for dynamic studies. Development of this pulse sequence was an important step in the success of this project. Excitation was performed using a dual tuned surface coil. Readout was performed using a trapezoidal fly-back echo planar encoding (Cunningham et al. 2005). EPSI unlike PRESS/STEAM CSI utilises only one slice selective RF pulse, and the localization in the spatial dimension is performed by a sweeping magnetic gradient i.e. kx . EPSI essentially accelerates the acquisition of spectral and spatial information. Consider a scan requiring 16x16 spatial points and 512 spectral points, EPSI will acquire all of the data in 16x1 TR steps, conventional chemical shift imaging techniques used clinically will acquire the same amount of data in 16x16 TR Automatic prescan is not available for multinuclear spectroscopy (on clinical MRI scanners). Auto prescan performs automated setting of appropriate recieve/transmit gains, tuning to a specific resonance peak and magnetic field shimming. Therefore the software to automate prescan was developed for a clinical MR scanner. This project required design of RF coils, and the multi tuned coil approach was chosen as it allows for simultaneous data acquisition and decoupling. Dynamic MRS studies were performed on human volunteers and standardized phantoms to calibrate and observe changes during pulse sequence development. Proton decoupling of 31 P data is critical for obtaining good quality brain spectra. This was done using a WALTZ-4 decoupling system during 31 P spectroscopy. However 23 PhD Thesis Sergei Obruchkov Dept. of Medical Physics much manual calibration is still required in order to perform decoupling experiments. Since the motivation behind EPSI was to reduce scan time then ideally it should not take more time setting up the patient and calibrating the equipment. Data analysis must be performed on the acquired spectra. While there are a number of software packages for analysing spectroscopy data (e.g. VARPRO, LCmodel, SAGE, MRUI SiTools, etc.), collected data must be imported from raw format into these packages first. A combination of SAGE and MATLAB software was used to read, reformat and process the data. This project was designed towards pre–clinical technical development of software and hardware as clinical scanners are ill equipped to perform 31 P MRS “out of the box”. Following technical developments tests were performed on a series of specially designed phantoms. The first in vivo human tests were performed on leg muscles of healthy volunteers during plantar flexion exercise inside the MRI. The technical developments can be translated to clinical settings including the studying of muscle diseases and breast cancer. St. Joseph’s Healthcare is a leader in MRI assessment of breast cancer and currently owns a state of the art MRI compatible breast imaging and biopsy system (Sentinelle Medical, Toronto Ontario Canada). Coupling rapid spectroscopy with breast biopsy data is ideal for studying breast cancer. Current MRI literature of 31 P spectroscopy presents clear evidence of increased concentrations of phosphomonoesters (PMEs), 24 PhD Thesis Sergei Obruchkov Dept. of Medical Physics and phosphodiester (PDEs) as well as inorganic phosphate peaks in breast tumour tissue (Leach et al. 1998; Arias-Mendoza et al. 2004). 25 PhD Thesis Sergei Obruchkov 26 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 2 Theory 2.1 Basic NMR Physics Nuclear magnetic resonance is a science of understanding the behavior of spin in a static magnetic field and its interactions with surrounding chemical and physical environments. The physical nature of spin however is a complex question that lies in the realm of theoretical physics which, along with other phenomena such as the nature of mass, inertia and gravity are still not very well understood. Uncovering the true nature of spin is not the subject of this thesis. However one point that must be left with the reader is that spin can be regarded as a fundamental property of the particle like mass or charge. In order to understand NMR principles an understanding of nuclear interactions within a magnetic field is required. First the energy of a proton in a magnetic field is considered, then the classical picture of a proton as a charged spinning particle is presented. 2.1.1 Quantum Description Once a proton with spin I, is placed in a uniform magnetic field B~a it possesses a ~ Magnetic moment and angular momentum moment µ ~ and an angular momentum ~I. 27 PhD Thesis Sergei Obruchkov Dept. of Medical Physics quantities are parallel vectors and they may be written as: µ ~ = γ~I~ (2.1) with γ as the magnetogyric ratio, a nuclear specific constant present to make the two quantities equal. For the hydrogen nucleus, or the proton, the magnetic moment is tiny 1.4102 × 10−26 JT−1 , and magnetogyric ratio is γ 2π = 42.575 MHzT−1 . In the applied magnetic field B~a , the energy of interaction for a particle with moment µ ~ is: ~ a. E = −~µ · B (2.2) The applied magnetic field is usually a vector in a single direction B~a = Bo ẑ then equation 2.2 may be presented as: E = −µz · Bo = −γ~Bo Iz (2.3) Where the allowed values of spin Iz are mI = I, I − 1, ..., −I which correspond to quantized energy levels EI = −mI γ~Bo . The number of quantized energy levels in the magnetic field is 2I + 1. This means that once a nucleus with non zero spin is placed in a magnetic field it can take on one of 2I + 1 energy states. This is known as Zeeman splitting and is illustrated in Figure 2.1 for hydrogen nuclei. 28 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.1: Zeeman splitting of energy levels in the presence of the magnetic field Bo . Higher energy state corresponds to those nuclei that are aligned antiparallel to Bo . Energy difference between two states is described by Eq. 2.4. The 1 H and 31 P nuclei in a magnetic field Bo have two possible energy states mI = ± 12 . The energy difference between the two states in terms of ~ωo can be calculated from equation 2.3 to be: ∆E = 2µBo = γ~Bo = ~ωo (2.4) NMR is a quantum mechanical phenomenon and the observed signal can be viewed as the energy measured between the transition of the proton from the higher energy state m− to the lower m+ . With a sufficiently large magnetic field, ∆E becomes larger than the thermal energy kb T at room temperature so there are no spontaneous transitions from lower to higher energy state. Most commonly this energy is in the form of electromagnetic radiation. Calorimetric techniques had also been attempted in order to measure magnetic resonance by Gorter (1936) however these methods did not succeed. 29 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.2: A spin placed in magnetic field Bo will precess around the field with frequency ωo . 2.1.2 Classical Description In a classical picture, rotation of the proton generates a magnetic moment µ ~ . In the absence of a magnetic field it is oriented randomly but when placed in a static ~o it aligns with it, either parallel or anti–parallel. Once aligned it magnetic field B ~o as shown in Figure 2.2; this is similar to the spin top will start to precess about B precessing in earth’s gravitational field. This is readily observed from the rate of change of angular momentum of the system, which is equal to the torque ~τ that acts on the system. Mathematically this is a cross product between the two vectors: ~τ = ~ dI~ ~o =µ ~ ×B dt (2.5) substituting from Eq. 2.1 this becomes: d~µ ~o . = γ~µ × B dt 30 (2.6) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Combining constants from the above equation into a single constant, ωo = γBo (2.7) where ωo is a natural frequency of the precession also known as the Larmor frequency. The solution to the differential equation Eq. 2.6 is the motion of the proton around the magnetic field, written in terms of ωo as: µ = µo e−iωo t (2.8) This precession is field dependent and further it dictates the conditions at which the occurrence of resonance is observed. If a proton precesses in a magnetic field with frequency ωo , and is irradiated with electromagnetic radiation field of the same frequency perpendicular to the Bo , the two fields couple, energy is absorbed and the spin is tipped. 2.1.3 Bulk Magnetization NMR experiments are performed on large numbers of spins; usually on the order of ~ can be viewed as the Avogadro’s number (6.0223x1023 ). Then bulk magnetization M sum of all the nuclei moments µ ~ in the unit volume: ~ = M X i 31 ~i M (2.9) PhD Thesis Sergei Obruchkov Dept. of Medical Physics It is possible to rewrite the equation of motion Eq. 2.6 using Eq. 2.9 ~ dM ~o = γ~µ × B dt (2.10) In thermal equilibrium at temperature T the magnetization will be along the ẑ direction: Mxy = 0; Mz = Mo (2.11) where Mxy is transverse and Mz is longitudinal magnetization. If nuclei are placed in a static magnetic field, and are in thermal equilibrium, the population ratio of spins in different energy states is determined by the Boltzmann factor for the energy difference of 2µBo : N+ − 2µBo = e kB T N− (2.12) where kB is Boltzmann’s constant (1.38065x1024 JK−1 ) and T is absolute temperature in Kelvin. The magnetization of a system is related to the population difference N+ − N− of the spins. The excess of spins aligned with the magnetic field contribute ~ z = (N+ − N− )~µ as shown in to the total nuclear magnetization of the sample M Figure 2.3. 32 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.3: In a static magnetic field the population of nuclei aligned with Bo is larger then those that are antiparallel to the field N+ > N− . This excess of spins with individual magnetic moments µ contribute to the longitudinal magnetization Mo of the entire sample. 2.1.4 NMR Signal Detection In order to measure the NMR signal, the longitudinal magnetization Mz must be tipped into the transverse x–y plane, which gives rise to transverse magnetization Mxy . This is achieved by applying a pulsed electromagnetic field B~1 perpendicular ~o at the resonant frequency ωo . In effect the B1 applies to the main magnetic field B ~ by a prescribed angle α. The torque torque onto the nuclear spins which rotates the M is dependent on the strength of the applied field B1 and the pulse duration. This can be represented on a Bloch sphere in laboratory coordinates, or in rotating frame of reference where the x–y plane rotates at the Larmor frequency (Figure 2.4). The B1 field is usually generated by a coil that is perpendicular to Bo . That coil can also be used to detect a time varying magnetic field created by transverse magnetization Mxy rotating at the Larmor frequency. 33 PhD Thesis Sergei Obruchkov Dept. of Medical Physics ~ z is flipped by angle alpha using the field B1 . In the Figure 2.4: Magnetization M lab frame the motion of the nuclei is described by a nutation around the z axis. In the rotating frame of reference the x–y plane is rotating at ωo and magnetization, in this example, is seen as tipping along the y’–axis. The rate of tipping is determined by the frequency and amplitude of the applied field B1 . The nuclei will absorb energy most efficiently at the Larmor frequency which from Eq. 2.7 is magnetic field dependent. When a sample is irradiated with RF satisfying ~ will tip away from the resonance condition in the rotating frame of reference, the M the z axis at a precessional frequency ω1 (Figure 2.4) that is dependent on the B1 RF field as described by: ω1 = γB1 (2.13) The magnetization will continue tipping as long as B1 is applied. Usually B1 is ~ by a specific angle α. Because the receiver applied in pulses of RF energy, tipping M is sensitive to magnetization components in the x–y plane the signal will be directly related to Mxy . The maximum signal is then obtained when a tip angle α = 90o . In order to detect NMR signals most efficiently a series of pulses must be applied (discussed bellow). 34 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Ignoring relaxation (i.e. return to equilibrium) effects , the measured signal S(t) from a 3D sample located in a static magnetic field Bo , following a B1 excitation pulse, with the distribution of transverse magnetization Mxy (~r) would be: Z S(t) = Mxy (~r)e−iωo t d3~r. (2.14) vol In this case it is not possible to distinguish signals originating from different locations within the volume. Instead, NMR averages signal from the entire sample. 2.1.5 Longitudinal and Transverse Relaxation Following the tipping of magnetization, by angle α, the resultant initial vector can be decomposed into two magnetization components: Mxy transverse and Mz longitudinal magnetization. Following excitation, the transverse component of the magnetization decays away while the longitudinal component returns to its thermal equilibrium state. The magnetization component Mz of the system approaches equilibrium magnetization at a rate proportional to the departure from equilibrium Mo . Mathematically this can be described by: dMz Mo − Mz = dt T1 35 (2.15) PhD Thesis Sergei Obruchkov Dept. of Medical Physics which is the differential equation for T1 relaxation. The solution to this is: Mz = Mo + (Mz (0) − Mo )e−t/T1 , (2.16) where T1 is called the longitudinal or spin–lattice relaxation time constant, and characterizes the return of magnetization to equilibrium along the z direction (Figure 2.5). This relaxation is done predominantly via exchange of energy between the nucleus and surrounding environment. The T1 values lengthen with increasing field strength, because the amount of energy nuclei have to release, in order to return to equilibrium, increases with Bo . At the atomic level the nuclei are surrounded by randomly fluctuating magnetic fields. These fields result from moving molecules and other nuclei all possessing magnetic moments. These fluctuating fields tend to enhance the energy exchange between the nuclei and the lattice, shortening T1 . This effect is the basic principle behind MR contrast agents such as Gd–DTPA–BMA. The behavior of the transverse magnetization component is dominated by the de– phasing of individual magnetic moments leading to magnetization loss (incoherence). This behaviour can be described by: dMxy Mxy =− dt T2 36 (2.17) PhD Thesis Sergei Obruchkov Dept. of Medical Physics which is the differential equation describing T2 relaxation. The solution to this is simply: Mxy = Mxy (0)e−t/T2 , (2.18) where T2 is known as the spin–spin or transverse relaxation time constant. The interaction of nuclear dipoles with each other causes some spins to precess faster or slower then others. This brings a uniform distribution of spins that causes loss of spin coherence and results in decay of transverse magnetization, as shown in Figure 2.5. Transverse relaxation is determined by molecular mobility where spins bound to big, slow–turning macromolecules experience rapid T2 decay while smaller more mobile spins exhibit much slower decay. Compared to longitudinal relaxation, transverse magnetization always decays faster, generally, in biological systems, T2 6 T1 In addition to T2 another source of loss of transverse magnetization is Bo inhomogeneities which give rise to a time constant T2∗ . Mathematically this is expressed as: 1 1 = + γ∆Bo ∗ T2 T2 (2.19) where ∆Bo defines the inhomogeneity of the external magnetic field (γ∆ Bo is sometimes called 1/T 2 0 ). In a biological tissue with nonzero ∆Bo transverse magnetization 37 PhD Thesis Sergei Obruchkov Dept. of Medical Physics is largely described by T2∗ . In most cases T2∗ ≤ T2 causing a shorter free induction decay (FID) and accelerated loss of NMR signal. However using special techniques such as the spin echo, the signal loss can be recovered. While T1 relaxation is an irreversible process, while transverse relaxation, T2 (arising from spin-spin interaction) on the other hand can be partially reversed to produce an echo. The phase coherence loss due to spin-spin relaxation is essentially irreversible, whereas the coherence loss due to magnetic field inhomogeneity,T2∗ , can be reversed by generating a so-called Hahn-echo (or spin-echo). This can be achieved by application of RF energy in order to re–phase the spins back into coherence. For example, consider a set of spins which at t=0 are all aligned in phase (Figure 2.6). The existing field inhomogeneities between the spins will create slightly different rates of precession thus the phase between the spins will accrue at different rates. For a set of spins the phase difference at location ~r and time t can be calculated using: φ(~r) = γ∆B0 (r)t (2.20) where ∆B0 (~r) is the magnetic field inhomogeneity at ~r. This phase accrual can be reversed by applying a 180o RF pulse at time t = T E/2. This will cause the spins to be mirrored in z such that the phase +φ(~r) will become −φ(~r). Since they are still located at their respective location ~r they will preccess at the same rates as before and at time t = T E will re–phase forming an echo, see Figure 2.6. 38 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.5: Longitudinal relaxation must recover to the initial state along z, as described by Eq. 2.16. The energy during the longitudinal recovery is absorbed by the surrounding lattice. Transverse relaxation occurs due to spin–spin interaction and acts to disperse the distribution of precessional frequency ωo . Some spins slow down while others precess more rapidly. The end result is loss of spin precessional coherence. Transverse relaxation is governed by Eq. 2.18. 39 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.6: Echo formation following a spin echo experiment. The transverse magnetization is flipped around the axis, while still traveling at the same rate all the spins reffocus to produce a spin echo at time TE. 40 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 2.2 Chemical Shift It is often impractical to look at a single, or even a small group of nuclei, alone. In the real world these events are too quick to measure or are in very hostile environments. Nuclei typically are surrounded by electrons and form atoms which in turn often form complex molecules. By taking a look at a single nucleus and an electron one quickly realizes that one of the most prominent sources of the field inhomogeneities around the nucleus is the surrounding electron shell. When both are placed in a magnetic field B0 , the electron will oppose the applied field resulting in an effective field B: B = B0 (1 − σ) (2.21) where σ is called the shielding constant, and is dependent on the chemical environment i.e. surrounding electron shell. The nuclei will resonate at slightly different frequencies depending on their physical position on the molecule itself (Figure 2.7). Using equation 2.21 and equation 2.7 it is possible to quantify the expected frequency in radians per second to be: ω = γB0 (1 − σ) (2.22) However it is often more practical to use a unit–less scale to describe chemical shift delta such that it is independent of the applied magnetic field B0 : δ= ω − ωref 6 10 ωref 41 (2.23) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.7: Electrons surrounding the nucleus give rise to the chemical shift effect. This is demonstrated using the 1 H spectrum of lactate where the electronegative oxygen atoms shift electrons away from the protons thus giving rise to greater chemical shift. Thus the CH quartet is more de–shielded then the CH3 doublet in the above spectrum. where ωref is a Larmor frequency of a reference chemical. Therefore chemical shift of the nuclei dictates the peak position on the frequency axis. Using this information and the fact that the area under the peaks is proportional to the number of chemically equivalent nuclei residing at each location it is possible to describe chemical structures. 42 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.8: The SI spin system has four possible energy states which dictate possible transition states that could be detected by magnetic resonance experiment a) In the case of non–interacting spins there will be four possible transitions, with only two different energy differences. b) Interacting spins will split individual levels producing unique transisition energies. 2.3 J–Coupling Peak splitting is a phenomena that is usually the product of the interaction between two or more nuclei. The nuclei with magnetic moments generate magnetic fields which tend to influence the neighbouring spins directly through space in the form of dipolar coupling. In liquids molecular tumbling averages out these effects and no peak splitting is observed due to dipolar coupling. Peak splitting in liquids, however, can still be observed due to indirect coupling through the electron cloud of the chemical bonds this is called scalar coupling or J–coupling. Take spin system IS where both I and S possess magnetic moment for example 13 C and 1 H atoms and the precessional frequencies can be written down as: ωI = γI B0 (2.24) ωS = γS B0 . (2.25) 43 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Adopting |↑i as the low energy state and |↓i as the high energy state of the spin I and S it is possible to write down four possible combinations that will result in four possible energy states: |↑↑i, |↑↓i, |↓↑i, |↓↓i. If the spins are not interacting with each other then the transition energy can be calculated using eq. 2.4. The energy level diagram for such a system is shown in Figure 2.8 a). The transitions between levels in the non interacting nuclei will be symmetric: ∆E = ~ωS = hfS (2.26) ∆E = ~ωI = hfI . (2.27) In an NMR spectrum this will simply appear as single peaks occurring at the Larmor frequencies of fS and fI . However if scalar–coupling is present this will perturb the energy levels of the system as shown in Figure 2.8 b). This perturbation can be explained by the fact that due to Pauli exclusion principle the electrons occupying the same electron shell must be anti–parallel to each other, and thus through the hyperfine interaction between electrons and the nucleus the energy levels will be slightly altered from the case were no scalar–coupling effect is present. The energy levels will be altered proportional to J/4, where J is called a scalar coupling constant. The four possible transitions will no longer appear symmetric instead four unique transitions are observed: 44 PhD Thesis Sergei Obruchkov Dept. of Medical Physics ∆E = h(fS + J/2) ∆E = h(fS − J/2) ∆E = h(fI + J/2) ∆E = h(fI − J/2). This appears as peak spitting on the NMR spectrum (see Figure 2.9) with the peaks separated by J(Hz). The scalar–coupling interaction between heteronuclear nuclei can generate complex splitting in the spectra that can be determined by the method of successive splitting or mathematically using Pascal’s triangle. The relative peak areas of split peaks can also be determined from the Pascal’s triangle where the relationship between the areas is directly proportional to the number of the successive splittings that need to occur in order to produce that peak (see Figure 2.9). J–coupling provides another extremely valuable source of information for properly identifying the chemical compound, however for the in-vivo applications, where many metabolites can reside in the same location with overlapping peaks J–coupling can decrease the sensitivity of detection by decreasing the amplitude of the peaks, spreading the frequency information and contaminating information from other metabolites. This is often undesirable and is often thought to be controlled or removed if possible. A technique called decoupling is often used in high resolution NMR spectroscopy in 45 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.9: Simulated spectra of the J–coupling effects on the 13 C spectra, a carbon nuclei instead of producing a singlet produces a doublet at 0Hz when one 1 H is chemically bonded to it, a triplet at 225 Hz appears for the CH2 group and a quartet at 500 Hz for a CH3 order to simplify the spectra and increase the signal to noise via Nuclear Overhauser effect. 2.4 Decoupling In order to decrease the interaction between two or more magnetic dipoles, experiments that require simultaneous control of both nuclei are required. By affecting certain energy level transitions it is possible to eliminate coupling effect which are observed as peak splitting in the NMR spectra. These types of experiments are referred to as decoupling. It has been observed from experiments involving irradiation of nuclei at two frequencies simultaneosly, such that when the magnetic moments of the two nuclei diverge (become orthogonal), the effects of the scalar coupling disap46 PhD Thesis Sergei Obruchkov Dept. of Medical Physics pear. Thus keeping two magnetic moments orthogonal to each other it is possible to reduce the J–Coupling effect and eliminate the splitting of the spectra. There are many different decoupling methods, all of which utilise a second magnetic field B2 . This field is produced at RF frequency of the nucleus that is spin-spin coupled to the nucleus of interest. For example we are interested in acquiring 31 P spectra from methylphospohic acid P (OH)2 O(CH3 ), due to coupling between hydrogens in the hydroxyl group and the 31 P nucleus acquired spectra appears as a quadruplet. One type of decoupling method applies a continuous B2 field. As long as the field applied is at the resonant frequency of 1 H in the hydroxyl group the resulting 31 P spectra will collapse into a singlet, however if B2 is even slightly off resonance the decoupling effect will be greatly diminished. This type of decoupling belongs to the narrow band decoupling methods they are typically very inefficient, and have a very low figure of merit Ξ ≈ 0.01: Ξ= 2π∆F , γB2 (2.28) where ∆F is the effective decoupling bandwidth. Narrow band decoupling methods require large amounts of RF power that can cause sample and tissue heating. Another type of decoupling method utilises what is called cycles and super cycles in order to improve the efficiency while minimizing the RF power required. WALTZ-4 is a composite decoupling sequence consisting of multiple flip angles with different phase cycling that when experimentally applied have broad bandwidth that they could perform decoupling, the Ξ ≈ 1.0 for WALTZ–4. 47 PhD Thesis Sergei Obruchkov Dept. of Medical Physics WALTZ–4 sequence applies 4 cycles of RF pulses in increments of π, one WALTZ cycle would be (1 π2 , 2 π2 , 3 π2 ) eq. 2.29, at the end of the four cycles the spins are brought back into the Mz orientation, by proper phase cycling eq. 2.30. For 31 R = 90x 180−x 270x (2.29) W ALT Z − 4 = RRR̄R̄ (2.30) P–1 H decoupling WALTZ-4 has been extensively used as the decoupling se- quence of choice ((Barker et al. 2001; de Graaf 2005b; Widmaier et al. 1998)). 2.5 Introduction to Imaging: Hard vs Soft RF excitation Slice selective excitation is one of the most basic principles behind Magnetic Resonance Imaging (MRI). It is possible to excite a specific volume of spins and spatially encode it through the combined use of magnetic gradients and RF pulses. For example: a gradient Gz can be superimposed on top of the main magnetic field Bo , this creates a gradient of spins with spatial distribution of Larmor frequencies along the 48 PhD Thesis Sergei Obruchkov Dept. of Medical Physics z axis. In order to excite a slice of thickness ∆z an RF pulse with bandwidth of ∆ω must be applied to the system where: ∆ω = γGz ∆z (2.31) The only spins excited will be those falling within the bandwidth of the applied RF pulse. Spins residing outside of the RF pulse bandwidth will remain at equilibrium. A summary of this is shown in Figure 2.10. SINC or soft pulses have been a natural choice for slice selective excitation as they have narrow bandwidth and the Fourier transform of a SINC pulse is a rectangular function, an ideal slice profile. A typical SINC function is given by: B1 = ) sin( πt t0 ASIN C πt ≡ At 0 πt t0 −NL t0 ≤ t ≤ NR to 0 everywhere else (2.32) with A as the peak amplitude at time t = 0, and t0 being one half width of the largest central lobe of the sync function, NL and NR are number of zero crossings or one less number of side lobes of the SINC function. The bandwidth of the SINC function or the FWHM of its Fourier transform, can be approximated as ∆ω ≈ 49 1 t0 (2.33) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.10: A magnetic field gradient Gz distributes spin precessional frequencies ω(z) linearly throughout a sample, along z. A Fourier transform of a SINC pulse is a square function in the frequency domain. Applying a SINC pulse of bandwidth ∆ωo will excite only those spins that fall within ∆z. Unlike soft pulses, hard or rectangular pulses have much broader bandwidth and are applied without magnetic field gradients. Thus hard pulses will excite many spins simultaneously, assuming a homogeneous B0 magnetic field. Generally the bandwidth of a rectangular pulse will depend on its length. When considering a rectangular pulse, 1 if |t| > t )= RECT ( T /2 0 if |t| < T 2 T 2 the Fourier transform is naturally: F(RECT ) = T SIN C(πf T ) 50 (2.34) PhD Thesis Sergei Obruchkov Dept. of Medical Physics (i.e. a SIN C function) with F W HM ≈ 1.21/T where T is the pulse length. Thus the shorter the RF pulse the broader its excitation profile. Short pulses have many other benefits, as they excite the entire spin system rapidly, and cleanly (i.e. excited spins will be in phase in the transverse plane). Imaging can be performed using either of the pulses, however hardware limitations, safety, and many other factors have to be carefully considered for the type of imaging and experiments needed. Hard pulses can be used for performing inversion, and 3D readouts, while soft pulses are used for 2D readout, slice selection and spatial localization. The flip angle is easy to control and is dependent on the B1 RF field strength and pulse duration T as Z T θ=γ B1 (t)dt. (2.35) 0 MRI utilizes many types of RF pulses for different types of purposes, in the following sections, amplitude and frequency modulated RF pulses, and their uses in cases where B1 inhomogeneity can not produce desirable flip angles, are described. Consideration of Using Surface Coils RF coils are used to obtain MR signal, these are high Q resonating circuits tuned to a specific frequency, where the resonating frequency depends on the nuclear gyro51 PhD Thesis Sergei Obruchkov Dept. of Medical Physics magnetic ratio and the magnetic field strength. There are two types of coils: surface coils and volume coils. Volume coils are typically large assemblies in a Helmholtz or birdcage geometry, that can produce a uniform B1 RF field which is used to excite the spins. Surface coils are much more efficient at picking up signal close to the coil, but their effectiveness significantly decreases at about half the radius away from the coil. They are rarely used for excitation as they do not produce a uniform RF field (see Figure 2.11). Therefore current approaches in MR utilize an array of receiveonly surface coils that can be positioned close to the patients body (i.e. to ensure good coupling) to achieve a large volume coverage to receive MR signal. Volume coils are used for excitation purposes. When using arrays of surface coils for RF receiving a large volume coil (body coil) performs RF transmission. These types of coils are superior for imaging applications but for spectroscopy, surface arrays add unnecessary complexity as it becomes non–trivial to add multiple signals together. Specifically, phase correction and scaling factors during the adding process may affect quantification during the data analysis. Figure 2.11: RF Excitation profiles of a surface coil (left) and volume coil (right). The flip angle iso–lines clearly show a region of uniform excitation that can be achieved using birdcage geometry. In MRI volume coils are typically used for excitation purposes, and surface coils are used for signal recieving. 52 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Off–Resonance Effects and RF pulses The RF pulses described in the above section and the flip angle (eq: 2.35) assume a uniform B1 field acting on spin isochromats exactly on resonance: ωRF = ω0 = −γB0 There are a few ways of thinking about the off–resonance case when ωRF 6= ω0 ; in the ‘swing’ analogy the ‘pushes’ can be very small but if they are timed right they can really get the swing going. Or one can think of the transition probability of spins |+i → |−i in the case of a 180o pulse: the probability is highest when the RF is on resonance and diminishes as the |∆ω| increases. It is often useful to think about the off–resonance as a vector component ∆ω γ which will effectively tilt the RF B1 field out of the xy plane. This is known as the effective field Bef f (see figure 2.12) with magnitude: |Bef f | = q B12 + (∆ω/γ)2 2.6 Adiabatic experiments Thermodynamics of NMR 53 (2.36) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.12: The effective field tips the rotation axis and therefore the expected flip angle will be different than expected. What does NMR have to do with Thermodynamics? Spin temperature Ts probed by a magnetic field is slightly above the temperature of an experiment and is one of the single most important reasons why we can detect NMR signal. The temperature difference contributes to the separation in populations which we are probing. At equilibrium this population is governed by the Boltzmann distribution as: γ~B0 P− − = e kB Ts P+ (2.37) where P+ and P− are the populations of spins in higher |+i and lower |−i states. In thermodynamics the definition of an adiabatic process is one that keeps entropy constant. This is often experimentally achieved by changing one of the physical conditions very slowly. NMR experiments were first done using continuous RF radiation and the magnetic field was swept until the resonance condition was met. This in54 PhD Thesis Sergei Obruchkov Dept. of Medical Physics duced absorption of energy from the RF field was detected as a change in the coil induction. Earlier NMR experiments were attempted using mangetocaloric measurements. In these experiments it was thought that spin temperature could be measured directly. However, those experiments were spectacularly unsuccessful. With later advancements in technology, physicists cooling atoms below µK started thinking about cooling spins further using magnetocalorimetry. The basic principle behind magnetic spin cooling relies on the fact that sweeping a magnetic field can excite or saturate a system. If the sweeping process is very slow it does not matter about system initial equilibrium, the final equilibrium will be dependent only on the field and type of system. This allows one to ‘dial in’ spin temperature. Careful experiments have been performed which showed this cooling effect using adiabatic sweep of a magnetic field, a good example of an adiabatic reversible process. If a sample at temperature Ti and field Hi is placed in an apparatus in which the magnetic field is slowly ramped down, the system will end up at a new position, Tf and field Hf see figure 2.13. This effect is often used to cool systems to temperatures approaching zero degrees kelvin, and is often used in low temperature superconductors, in masers, to achieve Einstein–Bose condensate, and other low temperature physics. Adiabatic RF Pulses From the above sections, we can see the effect of Bef f on flip angle θ is quite pronounced when using surface coils, where B1 tends to vary as much as a factor of 10 55 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.13: By slowly varying the magnetic field it is possible to reduce spin temperature. This is a reversible adiabatic process. over the volume of interest. Adiabatic pulses are often used in these situations, as they do not require complex setup of using separate coils for transmit and receive. Typically an adiabatic RF pulse is decribed in the form: B1 (t) = A(t)e−iω1 (t)t (2.38) where A(t) is the envelope and ω1 (t) is the frequency sweep function. One can think of these pulses as composite pulses containing many hard pulses with varying amplitude and frequency. There are a few interesting properties of adiabatic RF pulses, that make them interesting for surface coil imaging. First of all they do not require multiple coil set up (figure 2.11), allowing surface coils to be used alone for transmitting and receiving signal. Furthermore, they can act as inversion pulses that are RF amplitude independent, and also be useful in frequency selection as well. 56 PhD Thesis Sergei Obruchkov It is best to introduce a reference frame ∗ Dept. of Medical Physics called the frequency rotating frame which will rotate at the frequency of the RF field, i.e. modulated by ω1 (t). Starting with ~ is the product of the the Bloch equations, in the laboratory frame, magnetization M ~ and interaction between two magnetic moments: magnetization vector from spins M, ~ the magnetic field be it the static magnetic or magnetic component of the RF B: ~ dM dt ! ~ ×M ~ = −γ B (2.39) Lab where ~ = B1 cos(ω0 t)î0 + B1 sin(ω0 t)î0 + B0 k̂0 B (2.40) and î0 , ĵ0 , k̂0 are fixed unit vectors in the lab frame. For convenience we’ll define the RF field only in the x direction: B1 = B1,x Magnetization in the laboratory frame rotates at ω0 and î = î0 cos(ω0 t) + ĵ0 sin(ω0 t) ĵ = −î0 sin(ω0 t) + ĵ0 cos(ω0 t) (2.41) k̂ = k̂0 ∗ Lab reference frame is the one that we would observe as bystanders. The spin would appear to nutate or precess around the magnetic field. The rotating reference frame is one where we would be rotating at frequency ω around the spin. In this frame spins would appear stationary until RF pulses are applied. 57 PhD Thesis Sergei Obruchkov Dept. of Medical Physics then ~ = B1 î + B0 k̂ B (2.42) ~ = Mx î + My ĵ + Mz k̂ M ~ Where Mx , My , Mz and î, ĵ, k̂ are time varying in the lab frame. Differentiating M with respect to time produces: ~ dM dt ! dMx dMy dMz dî dĵ dk̂ î + ĵ + k̂ + Mx + My + Mz dt dt dt dt dt dt ! ~ dî dĵ dk̂ dM + Mx + My + Mz = dt dt dt dt = Lab (2.43) Rot In the laboratory reference frame vectors î, ĵ are rotating unit vectors at the Larmor frequency ω0 and can be expressed as a cross product so ω~0 = ω0 k̂ and the time derivatives would then be: dî dĵ dk̂ = ω~0 × î; = ω~0 × ĵ; = ω~0 × k̂ dt dt dt substituting above we get ~ dM dt ! = ~ dM dt = ~ dM dt Lab ! + ω~0 × (Mx î + My ĵ + Mz k̂) !Rot (2.44) ~. + ω~0 × M Rot 58 PhD Thesis Sergei Obruchkov Dept. of Medical Physics In the rotating frame then we can rearrange the equation such that ~ dM dt ! = Rot ~ dM dt ! ~ − ω~0 × M Lab ~ ×M ~ − ω~0 × M ~ = −γ B (2.45) ~ − ω~0 ) × M ~ = (−γ B ~ ~ + ω~0 ) × M = −γ(B γ where the quantity in brackets is known as the effective field ~ + ω~0 B~eff = B γ (2.46) ω~0 = −γB0 k̂ (2.47) At the Larmor Frequency of the applied magnetic vector simplifies to ~ = B1 cos(ω0 t)î0 + B1 sin(ω0 t)ĵ0 + B0 k̂0 B (2.48) = B1 î + B0 k̂ and the effective field will be −γB ~ ef f = B1 î + B0 0 k̂ = B1 î B k̂ + γ 59 (2.49) PhD Thesis Sergei Obruchkov Dept. of Medical Physics The changing reference frame cancels the static B0 field in the rotating frame, which is convenient, when working with frequency modulated RF pulses. An RF pulse with a time dependent ω1 (t) component in the form of eq: 2.38, can now be easily applied. The equation for the effective field can be calculated in the rotating frame ω0 + ω1 (t): ~ = A(t)cosω1 (t)îω0 + A(t)sinω1 (t)ĵω0 + 0k̂ B (2.50) ω1 (t) k̂ B~eff ω0 +ω1 (t) = A(t)îω0 +ω1 (t) + γ (2.51) This will create a z magnetization component equal to ω1 (t) , γ creating an effective field Bef f with the magnitude: s B ef f = A2 (t) + ω1 (t) γ 2 (2.52) This effect is also known as spin–locking, because the magnetization is locked and precesses around the effective field (see figure 2.14). Adiabatic Condition Now without the effect of B0 , the magnetization will precess around the effective field ~ to the effective Bef f . It is possible by sweeping ω1 (t) to lock the magnetization M 60 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.14: Left: With the cancellation of B0 in the rotating reference frame, the spins precess about the effective field. Right: For a trajectory with a slow sweep rate the magnetization follows the effective field vector in a full adiabatic passage experiment. field, this is satisfied when the angle Φ of Bef f changes relatively slowly compared to its magnitude: dΦ γ Bef f dt (2.53) which is known as an adiabatic condition. It is often convenient to define an adiabatic factor: γ Bef f η = dΦ dt (2.54) To achieve the adiabatic condition one will have to satisfy η > 1. In order to understand the effects of adiabatic RF pulses it is useful to plot amplitude γA(t) vs the ω1 (t) frequency sweep function. This is best done considering the hyperbolic secant function as an example (amplitude and phase of B1 according the general 61 PhD Thesis Sergei Obruchkov Dept. of Medical Physics form are presented in eq: 2.38). Amplitude and phase for the hyperbolic secant are defined as: γA(t) = γA0 sech(βt) 1 ω(t) = −µβtanh(βt) γ (2.55) A plot of γA(t) vs γ1 ω(t) (see figure 2.15), provides visualization and understanding of adiabatic pulses. The hyperbolic secant is known as an inversion pulse or adiabatic full passage pulse as it will tip magnetization 180o . It is possible to design different kinds of pulses, such as adiabatic half passage. This can tip magnetization by 900 and thus be considered as an excitation pulse. It is even possible to create an arbitrary angle rotation pulse. One example would be B1 insensitive rotation or BIR–4. (The description of BIR–4 will follow at later date and will be in more detail).(Slichter 1991) 62 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.15: Representation of a hyperbolic secant pulse (top left) and its frequency ~ ef f field rotating modulating function (top right). Bottom left: with the effective B clockwise, the rotation is controlled by the ω1 (t) function. Bottom right: Adiabatic pulses are insensitive to B1 . The shaded contour describes the area where the adiabatic condition is satisfied, i.e. the magnetization vector will follow the effective field when the adiabatic condition is met. In this figure adiabacity is best in the middle but is poor for the low and high B1 values. 63 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 2.6.1 Spatial Encoding ~ Varying the magnetic Spatial encoding of a selected slice is done using gradients G. field in the spatial domain, distributes the Larmor frequency of the nuclei as a function of their spatial position ~r according to: ~ · ~r ω(~r) = ωo + γ G (2.56) By prescribing gradients at specific times, intervals, and magnitudes, in combination with RF pulses, the MR signal S(t) can be spatially encoded. Thus, with the application of gradients, the measured NMR signal can be rewritten from Eq. 2.14 using the spatially dependent Larmor frequency: Z S(t) = Mxy (~r)e −iωo t+γ Rt ~ )·~ G(τ rdτ 0 d3~r. (2.57) vol The above signal equation with spatial encoding parameter ~k(t): ~k(t) = γ 2π Zt ~ )dτ, G(τ (2.58) 0 can be further simplified by understanding that the signal S(t) is demodulated. Thus the S(t) can be rewritten without e−iωo t as a function of the spatial encoding parameter ~k(t) as: s(~k) = Z ~ Mxy (~r)e−i2πk·~r d3~r vol 64 (2.59) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Thus the signal is acquired over time, as the gradients are applied to encode the position of the changing ~k parameter. In a 2D case then gradients Gx and Gy can be applied such that to collect a two dimensional map of the spatial encoding parameters kx (t), ky (t) we can rewrite these as: γ kx (t) = 2π Zt Gx (τ )dτ 0 and, γ ky (t) = 2π Zt Gy (τ )dτ (2.60) 0 After time=t the collected signal can be represented as a function of kx and ky : Z Z s(kx , ky ) = x Mxy (x, y)e−i2π[kx x+ky y] dxdy (2.61) y Therefore, from this equation the magnetization distribution Mxy (x, y) can be recovered via Fourier transformation. In general, the Field of View (FOV) associated with other modalities, is directly associated with the area that is imaged, or sampled. In MR, sampling takes place in the spatial frequency domain. In order to get the image the data is Fourier transformed: this transformation results in an inverse proportional relationship between the sampling rates ∆kx and ∆ky and FOV: F OVx = 1 1 , F OVy = ∆kx ∆ky 65 (2.62) PhD Thesis Sergei Obruchkov Dept. of Medical Physics The signal FID or an echo is acquired digitally. A 2D array of s(kx , ky ) or k–space data is produced by frequency and phase encoding. During frequency encoding a gradient, Gx , is applied during the read–out of the echo signal. This fills one line of k–space and resolves one spatial dimension x. To produce a second dimension a phase encoding gradient, Gy ,is applied with different strength. The two steps are repeated with varying amplitudes of Gy until all of the desired k–space is acquired (see Figure 2.16). The collected k–space data is digitized into real and imaginary components which are acquired by mixing the signal with cos(ωo t) and sin(ωo t) functions. The resultant real and imaginary k–space is converted to image space using a 2D Fourier transform. Often the number of points in k–space is acquired with the dimensions as a function of power of 2 to allow use of the Fast Fourier Transform (FFT) algorithm. 66 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 2.16: By applying Gx and Gy gradients it is possible to acquire 2 dimensional information about the distribution of Mxy (x, y). Steps 1 through 4 show the application of gradients and their direct effect of traversal through k–space. Data is recorded only during steps 2 and 4. Data collected in the frequency domain must be reconstructed using a 2D FT algorithm to produce an anatomical image. 67 PhD Thesis Sergei Obruchkov 68 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 3 Echo Planar Spectroscopic Imaging of Phosphorus and Hydrogen using Flyback Echo Planar Readout Trajectory. This chapter deals with the development and design of a MR pulse sequence which is used to accelerate Chemical Shift Imaging using a modified readout technique. Theory, of the design of echo planar spectroscopic imaging with flyback trajectory for use on 31 P nucleus is discussed. Results of EPSI sequence implementation on a clinical GE 3T scanner are presented. 3.1 Introduction 1 H magnetic resonance spectroscopic imaging (MRSI) or chemical shift imaging (CSI) is widely used as part of clinical protocols for measuring spatial distribution of different metabolites, (Kurhanewicz et al. 2000). However due to long acquisition times CSI is rarely used in a clinical setting. The commercial success of MR scanners with higher magnetic fields (i.e >3.0T), over the past few years was largely driven by the fields of neuroimaging and angiography (Krüger et al. 2001). The MR spectroscopy has also seen some gains by the increased signal-to-noise ratio (SNR) and narrowing of spectral resolution (Hetherington et al. 1997). However, theoretical predictions of two fold increase in SNR with the doubling 69 PhD Thesis Sergei Obruchkov Dept. of Medical Physics of the B0 , have not been observed (Gonen et al. 2001). In the field of spectroscopy even a slight increase in SNR can be traded to reduce the long scan times (Dydak and Schar 2006) according to the equation: SN R ∝ V √ t, (3.1) where V is voxel size and t is acquisition time. The 31 P MRS on the other hand has always been of high interest to researchers and clinicians giving the ability to measure energetic metabolism in vivo. However due to the lack of clinical scanner capabilities, 31 P MRS is mainly used for research purpose. In order to push the 31 P MRS into a clinical realm much technical development needs to occur. In this chapter the design of rapid localization techniques for 31 P spectroscopy is discussed. While there are many multi shot fast chemical shift imaging methods available for accelerating 31 P CSI (Pohmann et al. 1997), a recent study comparing fast spectro- scopic imaging methods by Zierhut et al. 2009 found that the best approach was EPSI, with a flyback gradient design, when compared to other k–space sampling methods. In contrast to parallel imaging acceleration techniques, and other rapid trajectories a flyback trajectory was found to be extremely robust in measuring and assessing metabolite concentrations in comparison to the gold standard CSI technique (Cunningham et al. 2005). 70 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Table 3.1: In vivo concentration of phosphorous metabolites in human brain tissue. Metabolite Adenosine triphosphate (ATP) Phosphocreatine, Inorganic phosphate (Pi) Concentration (mM/L) 3.03 5.18 1.5 For MRS the limit of detection is about 0.5 mM for metabolic concentrations, and 31 P containing metabolites have concentrations that are very close to this detection limit (Table 3.1) (Bottomley and Hardy 1989). Therefore the main goal of rapid imaging is to speed up the acquisition while keeping the signal to noise ratio for phosphorous metabolites in the range of SNR = 10 – 50 for valid data analysis. 3.2 Theory For gradient design hardware considerations, maximum amplitude and slew rate are important. This is often limited by the amplifiers, physical dimensions of the coils and the safety standards set by governing bodies to limit dB/dt exposure to the patients for prevention of peripheral nerve stimulation. 3.2.1 Chemical Shift Imaging signal equation As seen in chapter 2 magnetic resonance imaging usually deals with just relaxation parameters T1 , T2 and proton density ρ to describe the magnetization distribution m(x, y) and the effect of the imaging pulse sequence on the weighting it will produce. A fourth parameter, chemical shift σ, is often ignored and even considered an artifact 71 PhD Thesis Sergei Obruchkov Dept. of Medical Physics (chemical shift artifact), often requiring complex approaches trying to separate signal from fat and water (Merkle and Dale 2006; Dietrich et al. 2008). In order to understand chemical shift imaging we need to extend the magnetization description with frequency domain. For 2D CSI magnetization distribution can therefore be described using m(x, y, f ) . For example imaging muscle and fatty tissue in humans can be considered as imaging a volume from water H2 O and fatty acids CH2 , producing two lines separated by about 3.5 ppm. Most imaging methods integrate over the frequency domain, producing images that have equal contribution from fat and water. If scan parameters are not considered properly the signal from fat and water will appear overlapped resulting in a chemical shift artifact (called chemical shift artifact of the 1st kind, noted in the frequency encoding direction). Spectroscopic imaging methods, on the other hand, can be thought of as a combination spatial–spectral imaging method, where the goal is to obtain NMR spectra for each spatial position. This is often done at coarse spatial and high spectral resolution, due to low concentrations of metabolites. Signal from volume m(x, y, f ) can be written down as: Z Z Z S(t) = x y m(x, y, f )e−i2πkx (t)x e−i2πky (t)y e−i2πf t df dydx f 72 (3.2) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Table 3.2: SNR is proportional to square root of total acquisition time Acquisition Time Single Voxel MRS Number of Acquisitions 2D CSI NEX × X encodes × Y encodes 3D CSI NEX × X encodes × Y encodes × Z encodes which is essentially a 3D Fourier transform of the m(x, y, f ), where the kx (t) and ky (t) are trajectories through k-space and are functions of applied magnetic gradients: t γ kx (t) = 2π Z γ ky (t) = 2π Z Gx (τ )dτ 0 t Gy (τ )dτ (3.3) 0 Signal to noise ratio of 1 H MRS varies with the square root of acquisition time. Compared to single voxel spectroscopy, CSI has the same SNR as single voxel data and provides spatial information. (Table 3.2) shows how time acquisition depends on number of phase encoding steps. 2D CSI has 2 spatial and 1 spectral dimension. The simplest pulse sequence (i.e. a CSI sequence) for this type of acquisition scheme is shown in Figure(3.1 & 3.2). Due to the limited number of phase–encoding steps and large voxel sizes used in 31 P CSI (>1 mL), it is therefore important to consider the point spread function (PSF) of the pulse sequence. The shape of the PSF depends on the acquisition scheme, the 73 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.1: 2D CSI sequence and its associated k-space trajectory Figure 3.2: 2D CSI data sets showing both kx and ky spatial dimensions. coarse resolution and large voxel sizes can cause the possible side lobes of the PSF to cover substantial distances, thereby spreading incorrectly localized signal. Furthermore, nominal voxel sizes usually differ substantially from effective voxel sizes, as defined by the FWHM of the PSF. This becomes especially important in the design of fast CSI sequences, where sometimes the SNR per voxel is artificially increased by the large effective voxel sizes. Depending on the acquisition scheme it is often difficult to calculate and can be even more difficult to measure the real PSF of the sequence 74 PhD Thesis Sergei Obruchkov Dept. of Medical Physics (Robson et al. 1997). When using surface coils for excitation, measurement of the PSF becomes even more challenging as one needs to compensate for B1 inhomogeneities (Murphy-Boesch et al. 1998). The PSF can be calculated using the Fourier transform of the acquisition pattern in k–space, since EPSI reads out data in a Cartesian fashion the PSF of EPSI and the conventional CSI are very similar Figure 3.3. Figure 3.3: a) PSF of conventional CSI, which is the same as the flyback EPSI readout; and b) PSF of a circular k-space readout trajectory. The EPSI sequence with a flyback readout trajectory reads out data where the kx direction is traversed in an echo planar manner with readout only in one direction. This ensures that the phase of the spins always evolve in the same direction (Figure 3.4). This minimizes artifacts associated with the phase reversal in most echo planar spectroscopic and imaging techniques (Cunningham et al. 2005). 75 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.4: (Left) EPSI pulse sequence with a flyback readout trajectory. (Right) The readout is performed using Gx gradient showing the resulting k-space trajectory. 3.2.2 Optimization of Echo–Planar Flyback Readout Trajectory Gradients When designing an echo planar spectroscopic readout trajectory both spatial and spectral bandwidths must be considered (Figure 3.4). Spectral bandwidth can be defined as the repetition of individual segments of the flyback trajectory defined as: BWspect = 1 Tf + Tr (3.4) where Tf and Tr are the periods of the trajectory as shown in Figure 3.5. While the spatial bandwidth requirements are defined by the Nyquist limit in the field of view (FOV) direction, BWspatial = 1 γ Gx Lx 2 2π 76 (3.5) PhD Thesis Sergei Obruchkov Dept. of Medical Physics where Gx is gradient amplitude and Lx is the required FOV. The trajectory is further constrained by the MRI system itself, in particular the slew rate (SR) and maximum amplitude of the gradients and the maximum sampling rate. Therefore in order to design an optimal trajectory five parameters need to be specified: spectral BW, spatial resolution, maximum system slew rate, maximum gradient amplitude and the sampling rate. In order to calculate the optimal Gx gradient trajectory that will generate MR signal (i.e. an echo), with the required spatial bandwidth, the area under the readout gradient (|A|) must equal: |A| = 1 , γ δ 2π x (3.6) where δx is the required spatial resolution for a CSI voxel. Figure 3.5: Flyback EPSI trajectory broken apart into segments required for calculating needed parameters. The segment after the rewinder portion of the sequence is repeated the required number times. 77 PhD Thesis Sergei Obruchkov Dept. of Medical Physics If the desired spectral BW is evenly divisible by the sampling rate of the MR system, the problem simplifies to a geometric one where the area of any triangle can be defined using the system gradient slew rate as a constant (Figure3.5). Then defining the problem as a system of equations becomes easy: Tf + Tr = |A| + 2r + 2r1 h h = SR r (3.7) Tf + Tr = BWspectral The solution to the above system of equations was found to be: |A| 1 −r − 2 ∗ BWspect (2 ∗ SR ∗ r) 2 − |A| − r2 = 0 SR (3.8) where the spectral bandwidth, BWspect , gradient area, |A| and slew rate, SR are all known quantities. From the two solutions to the above quadratic equations the shortest r is chosen such that it does not violate the maximum gradient amplitude of the system. The above equation has no solution if the requirements for spectral and spatial resolution are too high i.e. when: (2.0 ∗ r − 1/BWspect ))2 − 4 ∗ |A| ) < 0. SR (3.9) In order to design the gradient trajectory the exact solution is linearly interpolated and converted to an even 16 bit signed integer vector with temporal resolution of 4 78 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.6: A Cartesian dataset is acquired using flyback echo planar trajectory thus maintaining an optimal point spread function, PSF. The EPSI data is slightly phase shifted in the kx direction. This is easy to correct by phase shifting the data set according to the acquisition parameters. µs i.e. the system gradient update time. Due to integer conversion the gradient area is verified to ensure read out and rewinder areas are exactly the same. Source code used for system design can be found in Appendix A. Efficiency of the EPSI readout is determined by its duty cycle (i.e. the time the sequence spends in the linear readout trapezoid Tf ) where the effective SNR ESN R can then be written down as: s ESN R = Tf Tf + Tr (3.10) The data that is acquired using an EPSI with a flyback trajectory does not exactly fall onto the Cartesian grid. Instead each point is shifted by a certain amount, φ, which depends on the spectral bandwidth and spatial resolution of the readout EPSI gradients see Figure 3.6. 79 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 3.2.3 Reconstruction The reconstruction of EPSI data can be performed in either the time or frequency domain however both methods first require reshaping the acquired time data into the 3D data set (i.e. 2 spatial, 1 spectral dimension) and then shifting each data point by the proper phase amount. Each 1D FID was first reshaped to form a 2D data set according to the acquisition parameters. For a 2D acquisition the result would be a 3D data set and for 3D CSI a 4D data set would be generated. The transformed data is Fourier transformed in the time domain creating a frequency domain data set I(f )mn were m is the sample number in the nth gradient echo. Recovering a Cartesian data set is then performed by applying a linear phase shift to each point according to equation: 0 Imn (f ) = I(f )mn ei2π mn∆tBWspect N (3.11) were m,n and ∆t are point location as described in figure 3.6 and N is the total number of gradient echoes. Equation 3.11 describes the phase shifting in the frequency domain, which in the time domain is equivalent to a sinc interpolation (aka Whittaker–Shannon interpolation) as described by ((Oppenheim et al. 1998)): x(t) = ∞ X x[n] · sinc n=−∞ 80 t − nT T (3.12) PhD Thesis Sergei Obruchkov Dept. of Medical Physics where T = 1/fs is the sampling interval, fs is the sampling rate, and sinc(x) is the normalized sinc function, sinc(x) = sin(πx) . πx (3.13) Following interpolation, the data is then Fourier transformed and is analyzed using tools designed for CSI analysis. The source code for the reconstruction algorithm can be found in Appendix B. Figure 3.7: The GE Healthcarer 3.0 Tesla MRI scanner in the Imaging Research Center of St Joseph’s Healthcare. 81 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 3.3 Methods and Materials A 3T Signa Excite HD MRI (GE Healthcarer Milwaukee, WI) equipped with the multinuclear accessory was used to perform all experiments. A 3.0 Tesla scanner is shown in Figure 3.7 with the specifications for the system provided in Table 3.3. The scanner has capabilities to perform imaging using two different gradients, whole and zoom, each with different slew rate and maximum amplitudes. Table 3.3: Parameters of MR Magnets used in the experiments. Magnetic Field Strength (T) Water Center Frequency (Hz) Maximum FOV (cm) Shimmed area (cm) Field Inhomogeneity (ppm) Band Width (Hz) 3.0 127,799,074 60 45x48 0.349593 500 Narrow Band RF Power Amplifier(Body Coil) Max Output Power (Watts) 35,000 Max Average Power (Watts) 1,250 Max Pulse Energy (Joules) 125 Max Pulse Width (ms) 50 Broad Band RF Power Amplifier(Body Coil) Max Output Power (Watts) 4,000 Gradients (Whole) Max Rise Time (µs) Max Amplitude (Gauss/cm) Slew rate (Tesla cm−1 s−1 ) 288 2.3 80 Gradients (Zoom) Max Rise Time (µs) Max Amplitude (Gauss/cm) Slew rate (Tesla cm−1 s−1 ) 268 4.0 150 Software OS version 12.0 M5B 0846.d 82 PhD Thesis Sergei Obruchkov Dept. of Medical Physics The pulse sequence was designed to use zoom gradients in order to gain high spectral and spatial resolution. Because the system has limited memory the trajectory was designed as two segments: a half rewinder and a trapezoidal readout with a flyback. When running this sequence it was important to ensure the designed segments were executed ’back to back’ with dead time minimized, in a periodic/symmetric fashion throughout the EPSI readout. This ensures symmetric k-space sampling and phase accrual. Design and optimization of the gradients was performed in MATLAB (The Mathworks, Natick, MA, USA). Gradient waveforms were then integrated into the scanner pulse sequence by modifying a CSI pulse sequence written in the EPIC (GE Healthcarer Milwaukee, WI) programming language. Acquired data was processed in MATLAB and formatted for use with the spectral analysis software SAGE (GE Healthcarer Milwaukee, WI). 3.4 Results Gradients were designed to ensure symmetrical data acquisition where dead time effects were taken into consideration in the acquisition scheme. Originally gradient files were designed to fill system machine memory limited to 16384 points and then to repeat itself, unfortunately this resulted in an 8µs dead time which shifted the k-space sampling every 16384 points. This occurred due to differences in the way the MRI system memory was managed, and the original gradient waveform file was 83 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.8: Instead of a linear readout the 8µs dead time caused phase accrual every 0.6ms resulting in a step like readout. designed. Therefore in this schema the acquired k–space readout could not be properly reconstructed since the points did not fall on a Cartesian grid, Figure 3.8. Dividing the gradient waveform into a rewinder and readout chunks ensured that gradients could be designed to be in sync with the data acquisition system, which has a limited set of readout filters with required bandwidths. The maximum readout bandwidth allowed by the system was BWreadout = 32.75 kHz. Therefore the designed gradients required BWspectral to divide evenly with the BWreadout . The collected data would then be acquired properly, Figure 3.9. Flyback echo planar trajectories were optimised for the available MR hardware. In Figure 3.10a), it is noted that data collected during the flat portion of the trapezoid contributes to the SNR, thus the EPSI efficiency will be higher for gradients designed 84 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.9: (Left) Designed gradients need to be equally divisible into the allowed system receiver bandwidth to ensure proper data acquisition. (Right) Gradients are out of sync with the receiver. for low spatial resolution and low spectral resolution, than those with higher resolution requirements. While 1 H MRS spans only 4 ppm at 127.8 MHz or 512 Hz, sampling at the rate of 1100 Hz is sufficient to satisfy the Nyquist criterion and as such is easy to design EPSI gradients at 1 cm. 31 P MR spectra, on the other hand, spans 25 ppm at 51.7 MHz or 1300 Hz, thus in order to acquire 31 P spectrum EPSI gradients need to have BWspectral = 2600 Hz. Designing gradients with high spatial resolution for 31 P becomes a real challenge. While any waveform gradient could be designed as long as they fall within the operating parameters of the MRI hardware, the gradients that were optimised and tested for 31 P are listed in Table 3.4. In order to have 1 cm and 2 cm spatial resolution, gradient waveforms were interleaved to ensure full spectral coverage (Figure 3.10b). 85 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.10: a) Waveform designed for 2cm spatial resolution and spectral bandwidth of 1000 Hz. Only the first 2.5 ms of the waveforms are shown. The calculated k-space trajectory and the points sampled at 31.25 kHz are shown as an overlay. b) In order to ensure complete spectral coverage interleaved waveforms were used to ensure full spectral coverage of the gradients waveforms. Table 3.4: List of readout gradients created and optimised for the 3T MRI system. 31 P waveforms Spatial Resolution (cm) 1 2 2.6 4 1 Spectral Resolution (Hz) 868 1302 1562.5 2083 ESN R % 73.9 68.8 63.5 64.3 504 93.0 H waveforms 1 3.4.1 Calibration Results To achieve the most accurate reconstruction possible the BWspect was measured directly from an EPSI sequence, for each optimised gradient, from a high concentration phantom producing high SNR data. This was done by applying gradients in an off resonance condition and observing phase of the FID signal (Figure 3.11), (Jung et al. 2007). 86 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.11: Measuring FID in an off resonance condition. (Left) the phase of the FID will reflect the k-space trajectory and its derivative will reflect the applied gradients (Right). After the excitation RF pulse, a gradient echo will form every time the gradients travel through the center of k-space. With an echo planar flyback trajectory a gradient echo is formed twice, once during the flat trapezoid gradient time Tf and once during the flyback Tr . Only the echo generated during the Tf is used during the reconstruction process. In order to verify that the sequence is operating properly and is recording data symmetrically, high SNR phantom data was used (Figure 3.12). Multiple echoes from raw data were superimposed on top of each other to display 256 gradient echoes. In figure 3.12b) the first points (1–55) are produced during the trapezoidal readout Tf , while the last points (55–63) are the echoes generated by the flyback gradients. Ensuring that the flyback echoes are positioned in a tight grouping, at the end of the dataset demonstrates quality of the readout gradient. 87 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.12: 256 Gradient echoes generated by the Echo Planar Readout spanning 63 points across k–x a) Raw data showing the echoes generated by the trapezoid and the flyback gradients b) After the data is properly reshaped to fit back on to the Cartesian grid the rewinder echoes are tightly bunched and are properly positioned at the end of the dataset. 3.4.2 EPSI Results Flyback EPSI and CSI data was acquired using gradients with 1 cm spatial resolution for both 31 P and 1 H. 31 P spectra were acquired using gradients designed with spectral bandwidth of 868 Hz (interleaved 3 times) and effective SNR of 73.9% (Table 3.4). 1 H spectra were acquired using gradients with BWspect = 504 Hz and ESN R = 93.0% as calculated per equation 3.10. A spherical phantom, filled with 0.3% saline (NaCl), was used to acquire 1 H CSI and EPSI data Figure 3.13. Chemical shift images were produced by integrating the area under the water peak. The resulting EPSI images geometrically correlated well with CSI data, with line widths (FWHM) of 30±5 and 25± 5 Hz respectively. 88 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.13: CSI and EPSI data of a Spherical Phantom (Top) 1 H 12x12 CSI data 1 cm3 spatial resolution 504 Hz spectral resolution no water suppression, scan time 4.8 minutes. (Bottom) 18x12 EPSI 1 H reconstructed data 1 cm3 504 Hz spectral resolution acquisition time 24 seconds. SNR comparison test of 31 P CSI and EPSI data were acquired on phantoms con- taining 1M concentration of KH2 P O4 and K2 HP O4 (Sigma-Aldrich, St. Louis, MO, USA) (Figure 3.14). CSI images were produced from computing areas under the two chemical peaks. EPSI offers acceleration times, for low bandwidth applications, 20 times faster then conventional CSI. Measured SNR varied according with theoretical prediction and decreased as a square root of the acquisition time. For 8 NEX and 2 NEX, SNR was measured to be 35% and 21.8% respectively, compared to that of the 89 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.14: High Resolution CSI data from 1M mono and di–basic potassium phosphate buffer solution. Images produced from the 20x20 CSI spectra and interpolated to 128x128 for display purposes. CSI acquisition. The 1 cm gradients corresponded well with theoretical predictions for the SNR Figure 3.15. 3.5 Discussion Successful design and implementation of 31 P EPSI sequence with flyback trajectory has been demonstrated. High SNR EPSI sequence with flyback trajectory can speed up acquisition without significantly affecting image quality as assessed using peak line width and geometric representation. 90 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 3.15: Relative SNR data as measured using ESN R = 73.9% gradients with 1 cm spatial resolution compares well with theoretical predictions. Overall EPSI data showed good correspondence with CSI data. Since flyback trajectory does not contain reversed echoes the quantification of spectra is more robust and is insensitive to acquisition errors. Thus this approach has great potential for in-vivo applications (Ulrich et al. 2007; Posse et al. 1994). 91 PhD Thesis Sergei Obruchkov 92 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 4 Design and Construction of a Heteronuclear 1H and 31P Double Tuned Coil for Breast Imaging and Spectroscopy This chapter deals with the development and design of a multi–tuned RF coil for simultaneous 31 P spectroscopy and 1 H WALTZ-4 decoupling. Background theory and design of a double tuned RF coil for breast imaging and spectroscopy application is presented. The results from RF calibration and imaging using the MRI coils are discussed. 4.1 Introduction The importance of RF coils was previously discussed in chapter 2. The MR signal is in a form of RF waves that belongs to very high frequency (VHF) band 30–300 MHz with typical bandwidth of less then 1 MHz. MR coils are designed to operate at a very specific frequency which depend on the Larmor frequency of the nucleus being imaged. Therefore a separate coil is needed for each nucleus of interest, e.g. one coil for 1 H and one for 31 P, 3 He, 13 C etc. However there are a specific type of coils called dual tuned or multituned coils that can operate at two or more frequencies simultaneously. This enables the MR scanner to perform experiments which require ability to interrogate multiple nuclei at the same time. 93 PhD Thesis Sergei Obruchkov Dept. of Medical Physics The practice of using volume coils for transmit and using surface coils for receive occurred in the last 15 years, and has played a large role in being able to increase the strength of the magnets while optimizing the available SNR. There are two main differences between coils that are used for transmission and reception of RF signal. Transmission coils are typically much larger and must be able to handle large amounts of power 15–35 kW is a typical strength of RF power amplifier that are used today in clinical imaging. Most clinical scanners have a built in transmit coil tuned to 1 H frequency called body coil, which is driven by a large narrow band power amplifier designed to run at the same frequency. Therefore a separate transmit coil and a broadband RF amplifier are needed for work with nuclei other than 1 H. Reception RF coils on the other hand are much smaller in size and use low power components. In order to use them safely the receive coils are detuned during the time when the transmit coils are turned on thuss isolating the coil. This is accomplished in two ways: passive and active blocking circuits. Both utilize diodes and capacitors to detune the center frequency such that the transmit and receive coils will not be interacting. T/R switches are typically used for these applications and often each T/R switch is designed to operate only at one frequency at a time therefore two T/R switches are needed to be used with a dual tuned coils. Multinuclear work requires many extra components that are often not available with most clinical scanners. When designing a dual tuned coil one must then consider multiple factors such as; type of experiments the coil will be used in, target anatomy, physical size, number of ports, transmit receive capabilities and many more. 94 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 4.1.1 Motivation for Dual Tuned Coil for Breast 31 P Spectroscopy With Two Port Input Normally if a scanner has multinuclear capabilities, in order to perform phosphorus spectroscopy or imaging only require a coil tuned to 31 P would be required. However, there are several common applications that utilize multiple–frequency excitation to study the interactions and to enrich acquired data with extra information. For example, it is possible to increase SNR by transferring spin polarization from 1 H to 13 C using the nuclear Overhauser effect (Bomsdorf et al. 1991). Proton decoupling is an effective technique for mitigating the effects of spin–spin coupling which tends to broaden and increase the complexity of the spectra (Ch 2). And the inverse detection methods used in 1 H and 13 C spectroscopy and many other application all require coils that can operate with high efficiency with two or more nuclei. In addition, dual tuned coils, are very useful for producing shimming and localizer/scout data using the 1 H channel prior to doing abundance of 31 P, 13 31 P spectroscopy. Because of the low C and other nuclei, 1 H shimming and localization often helps to increase the accuracy of spectral localization and the quality of field homogeneity. Furthermore the simultaneous acquisitions of 1 H and very useful, as it is possible to interleave 1 H and 31 31 P data has been shown to be P acquisitions (Jesmanowicz et al. 2002). Phosphorus spectroscopy presents clear evidence of increased concentrations of phosphomonoesters (PMEs), and phosphodiester (PDEs) as well as inorganic phosphate 95 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.1: a) Sentinelle Medical Vanguard breast biopsy table attached to a clinical scanner. b) Breast coils are swapable for a grid to MR guide breast biopsy c) Vanguard table with the MR coils. peaks in breast tumour tissue (Leach et al. 1998; Arias-Mendoza et al. 2004). Therefore it is of great interest to create and design a coil that would be compatible with clinical workflow and compliment the use of a specialized table that utilizes breast compression for imaging and performing breast biopsy (Figure 4.1). A double tuned coil, compatible with Sentinelle Vanguard breast biopsy table (Sentinelle Medical Inc., Toronto ON), was designed and built to efficiently operate at resonant frequencies of 31 P and 1 H. Dual tuned RF coil allows performance of exper- iments that simultaneously require RF transmission at both frequencies, as well as better quantification of 1 31 P metabolites, by applying B1 corrections acquired using H channel. 96 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 4.2 Theory This section will describe basic electromagnetic theory, from the creation of electromagnetic radiation to the basic principles behind RF circuitry and MRI RF system in general. An overview of modern design of MRI transmit and receive RF chain needed for implementation of dual tuned coils in MR experiments is provided. 4.2.1 Propagation of Electromagnetic Waves and Magnetic Field Generation In chapter 2 B1 field which is responsible for the generation of transverse magnetization Mxy and its decomposition into circular components have been discussed. The generation of the this time–varying magnetic field requires, like any electromagnetic field, some arrangement of electric charge. This is achieved using accelerating charges and changing currents. However unlike radio stations that use electrical dipole antennas to broadcast their signal with high efficiency over hundreds of kilometers, MRI systems require the use of magnetic dipole antenna to produce electromagnetic radiation, which are typically not very efficient. However they can produce good directionality and coverage over short distance e.g. human body. Take a wire loop of radius a, around which alternating current I is being driven as: I(t) = I0 cos(ω0 t). 97 (4.1) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.2: A circular loop with current I will generate magnetic field which could be calculated using Biot–Savart law. By driving the current I at resonant frequency of ω0 = γB0 two things occur: a B magnetic and E electric fields will be generated they will be perpendicular to each other and the coil and secondly they will radiate outward. The radiation carry energy away from the coil, typically electric field is responsible for radiating more energy then magnetic field, and the ratio of the amplitudes of the two components are E/B = c, where c is the speed of light. It is possible to calculate the field profile that is being radiated by an arbitrary shaped source. This is typically done using a time–variant solutions to Maxwell equations known as Jefimenko’s equations which describe generalized time–varying Coulomb and Biot–Savart’s laws (Griffiths and Inglefield 1999). Often the method of moments analysis subdivides a coil into segments from which, first the electrical field that is produced by the coil can be calculated 4.2, and later the magnetic field. 98 PhD Thesis Sergei Obruchkov ~ ~ = −∇V − ∂ A , E ∂t Dept. of Medical Physics ~ =∇×A ~ B (4.2) ~ and V~ are retarded potentials which can be calculated using: Where the A Z ~ ~0 J(r , tr ) )dτ 0 0 |r − r | Z 1 ρ(r~0 , tr ) 0 dτ V~ = 4π |r − r0 | ~ r, t) = µ0 A(~ 4π (4.3) (4.4) ~ r~0 , tr ) is the current density. A good review where ρ(r~0 , tr ) is the charge density and J( of method of moments analysis can be found at (Harrington 1993). This approach is often used with a simplification for coils that are typically smaller then the wavelength of the EM radiation emitted from them. While the exact solution to the magnetic field is complicated, it is possible to estimate it using Maxwell’s equations and static Biot–Savart law. Analytical solution can be separated for the distribution of magnetic B1 field from a single loop coil in radial and axial directions as: 99 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.3: Estimated magnetic field profile from a single loop coil, B1 profile of the XY plane at z=0. B1radial ∝ q x2 + z 2 (r + B1axial y √ r 2 + x2 + y 2 + z 2 2 −K(κ ) + (r − √x2 + z 2 )2 + y 2 E(κ ) x2 + z 2 )2 + y 2 (4.5) 2 r 2 − x2 − y 2 − z 2 2 2 √ ∝q K(κ ) + E(κ ) √ (r − x2 + z 2 ) + y 2 2 2 2 2 (r + x + z ) + y 1 (4.6) where K(κ2 ) and E(κ2 ) are Legendre’s elliptical integrals of the first and second kind, κ is an elliptical modulus with values between 0 < κ2 < 1 and r = x2 + y 2 + z 2 . The B1 field is visualized in Figure 4.3 (de Graaf 2007). 100 PhD Thesis Sergei Obruchkov Dept. of Medical Physics RF coils can be made out of large loops of copper however using a long continuous piece of copper or even multiple loops is not desirable. Large coils increase the size of the inductor, and because inductors are electrically lossy the image quality will be degraded. It is common practice to introduce breaks in the RF coil, thus separating long inductor into several components and distributing capacitance throughout the entire RF coil. Image quality will also largely depend on the scan parameters used to perform imaging experiment. For example Gradient Echo (GE) and Spin Echo (SE) will produce images with quite drastically different images. Typically when using GE based sequences such as spoiled gradient echo (SPGR) can produce images of better quality with transmit–receive surface coils then SE counterparts. This is largely due to the fact that transmit surface coils have hard time refocusing spins properly, and in cases where the TR << T1 , smaller flip angles is more desirable to maximise the SNR of the images (Evelhoch et al. 1984). Furthermore the Adiabatic RF pulses could be used to maximize the sensitivity of the surface transmit/receive coil, see Chapter 2. 4.2.2 Resonating RLC Circuits Any loop of wire can be used to transmit and receive RF, however the properties of that loop will define the optimal alternating current that can easily travel through that conductor. Typical MR coils can be simplified to RLC circuit which maximize the energy transfer at a given frequency. These can be found by solving Kirchoff’s 101 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.4: Resonating RLC circuit. law for RLC circuit shown in Figure 4.4. The sum of potentials across the resistor, inductor capacitor and source V must be equal to zero V : −L d2 Q dQ Q −R − +V =0 2 dt dt C (4.7) One can then solve the above differential equation to find out the maximum current that the circuit can support. However it is easier to understand the behaviour of the circuit from the phasor notations. Because capacitor and inductor both have a similar property such that current and voltage are π/2 out of phase with each other, it is often preferred to introduce the impedance of an element Z = V /I and write it such that the impedance for inductor and capacitor is in a complex plane i.e. orthogonal to the resistance. 102 PhD Thesis Sergei Obruchkov Dept. of Medical Physics ZR = R, ZL = iXL = iωL, ZC = iXC = − (4.8) i ωC Where the real impedance Real(Z) = R is called resistance and complex impedance Complex(Z) = X is called reactance. The inverse of an impedance is an admitance: Y = 1 . Z (4.9) Now the Kirchoff’s law still applies and one can write down the total impedance of the circuit shown in figure 4.4. −RI + i I − iωLI + V = 0 ωC (4.10) where ω is the angular frequency at which V is oscillating. The maximum current that one can generate can be calculated using I= Vo i(ωt−φ) e |Z| 103 PhD Thesis Sergei Obruchkov Dept. of Medical Physics where maximum current will be Vmax Imax = p R2 + (iωL − i/ωC)2 (4.11) The maximum current traveling through the circuit will depend on the impedance of the circuit Z and will be frequency dependent. The impedance can be modeled as: Z 2 = R2 + (jXL + jXC )2 Z 2 = R2 + (jωL − 1 ωC)2 (4.12) (4.13) The current within the circuit will be maximized when the impedance is minimized. This will occur when the reactance component of the impedance vanishes and this corresponds to frequency ω: 1 =0 ωC 1 ω = ω0 = √ , LC ωL − (4.14) (4.15) where ωo is known as natural resonant frequency of the circuit. Applying voltage at resonant frequency ω0 will maximize the current and thus the magnetic field B1 that is generated by the coil. The power transfered from the source to the coil itself will largely depend on impedances of the two systems. It is common to utilise 50 Ω transmission lines for their power transfer at broad range of frequencies. Thus the RLC circuit must be matched to 50 Ω. 104 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Tuning and Matching It is often not enough to just build a resonating circuit that can be used for MR imaging. The coil must be connected to the transmitter, receiver or the preamplifier. This is typically done though coaxial lines with impedance of 50 Ω. Thus the coil that is tuned and matched to 50Ω can theoretically achieve 100% power transfer and 100% efficiency simultaneously in a perfect resistance free reactive circuit. In an imperfect reactive circuit (i.e. one with both reactance and resistance), the efficiency η will be the power dissipated in the load PL divided by the power provided by the source PS . η= PL PS When the resistance is a small fraction of the reactance, the efficiency will be very high even at the maximum power transfer point. This is a key to designing high efficiency MR coils, ensuring that the current generated in the coil from the transverse magnetization of the spins will be detected with maximum efficiency. An introduction of a matching capacitor CM is necessary to convert the loaded coil impedance from usually a very high impedance to 50 Ohm, Figure 4.5. The natural frequency of the circuit will shift and thus the value of a tuning capacitor CT needs to be changed such that the coil will resonate at the needed frequency ω0 with the impedance at that frequency being 50 Ohm. 105 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.5: RLC circuit connected to the source ZS = 50Ω via coaxial cable Z0 = 50Ω. In order to ensure efficient power transfer at the resonant frequency tunning CT and matching CM capacitors are used to ensure that the coil acts as a 50Ω load at the resonating frequency ω0 . 4.2.3 Reciprocity Principle and Coil Quality Factor Reciprocity principle states that the B1 field generated by current It should be iden~ . In other words the tical to current Ir generated in the coil by magnetic moment M knowledge about the transmit field of the coil will give direct information about the signal generated in the coil. This symmetry between received and transmitted fields can be used to map the B1 , which in turn could be used for quantification of MR signal, e.g. molar concentrations of metabolites. It is possible to measure B1 field of the coil using several different methods, this information then can be used to compare a known concentrations of a reference sample to unknown concentration in the tissue. From section above we know that LCR circuit will resonate when reactance is zero, while the resistance of the coil usually peaks, Figure 4.6. Thus energy will be dissipated in the circuit decreasing the quality of the received signal. A quality factor Q of the built coil, is defined as the ratio of maximum energy stored in the circuit 106 PhD Thesis Sergei Obruchkov Dept. of Medical Physics over the total energy dissipated. Quality factor can be measure from the resistance of LCR circuit and it is related to the FWHM of its peak: Q= ω0 2∆ω (4.16) This quality factor depends on the design of the coil and its components. Quality factor of a tuned and matched unloaded RF coil can be approximated by Qunloaded = ωL , Rc where Rc is the resistance of the coil. RF coils ideally should have high Q so that they will be very sensitive to the RF frequency, without damping or dissipating RF energy. However as soon as the coil is placed on the sample or a patient that is magnetically lossy (sample usually introduce high resistance in the coil), the coil will couple to the sample and the Q will be reduced to: Qunloaded = ωL , Rc + Rs where Rs is resistance introduced by the sample. The coil sensitivity, S is defined as the ratio of quality factors of loaded to unloaded coil as S = S0 s 1− Qloaded , Qunloaded 107 (4.17) PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.6: Resistance and Reactance of the coil tuned and matched to 50 Ω. Resistance of the coil will change depending on the load that is placed onto the coil. Ideal col will have very narrow and tall peak thus have a large Q. where S0 is the theoretically maximum achievable sensitivity when Rc → 0. It is common practice to use a saline phantom with loading properties similar to human tissue for tuning and matching the coil. While the phantom can not mimic all of the loading characteristics that the device will be subjected to in clinical imaging it is often impractical to tune and match the coils for every single subject. 4.2.4 Transmit–Receive Switch The purpose of a transmit–receive (T/R) switch is to redirect RF, protect the receiver and the coil electronics. Additionally because most T/R switches use diodes, noise 108 PhD Thesis Sergei Obruchkov Dept. of Medical Physics between transmit and the receiver port is uncoupled ensuring proper operation of the low noise pre amplifier. The resonating circuits are very efficient at absorbing energy at their resonant frequencies. It is often dangerous to operate an unplugged coil in the bore of the magnet or have two transmit coils that resonate at the same frequencies operate in the magnet at the same time. During transmit the energy that is delivered to the coil often exceeds 2kW for small birdcage and linear coils, the body coil on the other hand often requires more than 10kW of power to generate required B1 field (Hayes 2009). The receive only coils are typically made of low power components that are not designed to handle large amount of power. Therefore a number of different techniques are utilized to ensure that the receive coil is detuned during the transmit phase. Typically transmit receive switches utilize DC bias and diodes to direct RF. Figure 4.7 shows two types of TR switches, Fig. 4.7a that can be used for simple linear coils, and Fig. 4.7b T/R switch for hybrid coils. During the transmit a forward DC bias is applied to the two diodes turning them on. Diode D1 lets RF through, and D2 shorts the λ/4 cable. From the coil view it becomes an open circuit, therefore no power goes through, thus protecting the preamplifier and the receiver. During receive phase, reverse bias is applied to the diodes, D1 effectively disconnects the transmit, and the coil is connected directly to the receiver port through the transmission line. During the whole process the matching of the coil remains the same to ensure most efficient power transfer. 109 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.7: a) Linear coils usually utilize a single port for transmission and reception. b) Hybrid coils actually utilize four ports, two of which are a separate transmit and receive ports. Because the quadrature hybrids are not always perfect, transmit and receive ports are separated to ensure proper operation. This type of a TR switch, is commonly used because it can handle both linear and quadrature coils. Quadrature hybrid is a 4 port device that during transmit splits RF power from port 1 equally into port 2 and 3. The RF in port 2 and 3 are π/2 out of phase with each. During receive mode the RF from port 2 and 3 are recombined into a single channel on port 4. Ideally port 1 and port 4 are completely isolated from each other so in principle no T/R switch is necessary. However, because it impossible to tune and match for every patient exactly, the reflected power will be recombined by quad hybrid into port 4, thus a T/R switch is still required, see Figure 4.7. During the transmit forward DC bias is on, RF is transfered into the coil any reflected power will be sinked into the load on port 4. During receive reverse biased is applied, the 110 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.8: a) Typical trap circuits used to split the resonances in a dual tuned coil design b) This approach uses DC bias and extra capacitors for dual tuned coil design. transmit port is effectively terminated by a 50 Ω load, and the RF can enter the receive port through the λ/4 transmission line (Chen, Hoult, and Sank 1983). 4.2.5 Methods for Dual–Tuning Coils Currently there is a great demand on using magnetic resonance imaging and spectroscopy for multinuclear applications. These demands often require the ability to perform 1 H localization, B0 shimming and proton decoupling which could be achieved using dual tuned probes. The dual tuned probes could be subdivided into two categories: single coil and dual coil approaches. Single coil rely on introduction of extra lumped components to the MR coil in order to increase the number of resonances. The dual coil approach uses multiple coils and separates the two modes using either geometric decoupling or using trap circuits. Generally the application in which the coil will be used is the main constraint for the design of a dual tuned coil. 111 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Single–Coil Method From section 4.2.2 we know that at the frequency when reactance of the coil reaches zero the coil resonates. So it is possible to design a circuit such that reactance goes through zero at multiple frequencies. There are a few ways of achieving this either using trap circuits or having extra capacitors that are activated using DC bias. Trap circuits are a type of tank circuits typically build out of capacitors and inductors in parallel. A dual tuned circuit introduce by Schnall et al. 1985 uses these tank circuits (Figure 4.8a) to create multiple resonance modes. These trap circuits are purely reactive components made up of capacitors and inductors in parallel, they can be used to design coils with any number of resonances. Typically, however, a single coil is not tuned to more then three resonances. The trap circuits are often non ideal and always have resistance associated with them. This often causes dual tuned coils to have lower Q then the single tuned counterparts. As well, often it is desirable to have multiple ports for the coil and each port must be tuned and matched to its particular frequency, this is often is the most challenging part of building dual tuned coils. This approach has been used to design surface or volume coils by Murphy-Boesch et al. 1994 and Matson et al. 1999. Another approach uses diodes and DC bias to electrically connect a second capacitor to the circuit (Figure 4.8) such that the resonant frequency of the coil will change. This approach however can utilise only a single frequency at a time (Weyers 2009). The advantages of this design is the fact that the coil can be optimized for two frequencies but they cannot be used at the same time. 112 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.9: a) Two coils are tuned and matched separately for the two nuclei of interest while physically they are super positioned on top of each other. b) A schematic diagram showing the approximation of the two coil approach. The inductive coupling of the two coils acts similar to a trap circuit splitting the resonance in two. Dual–Coil Method Crossed Elements Configuration It is possible to use two surface coils tuned to two different frequencies of interest and placed adjacent to each other, Figure 4.9. This approach is similar to using trapped circuits as the coils will be inductively coupled to each other, shifting the desired frequencies. Once the coils are placed next to each other and loaded with a sample the coils are retuned and rematched to the final desired frequencies. Similar to the trapped circuits when operating the system at the hydrogen frequency, the phosphorus coil will act as an inductor; while at the phosphorus frequency, hydrogen coil will act as a capacitor (Fitzsimmons et al. 1993; Fitzsimmons et al. 1987). This approach is very useful as it is the easiest implementation of two port dual tuned coil. It is often used when the design permits the position of two coils close together and if the coupling effects can be geometrically eliminated. Crossed element configuration of RF coils is often used as a form of geometric decoupling to fix the problem of counter–productive current generation. Generating two magnetic fields that are perfectly perpendicular to each other stops the flux from 113 PhD Thesis Sergei Obruchkov Dept. of Medical Physics inducing currents in the adjacent coil, Figure 4.10. Orthogonal configuration of the coils is often the simplest way of producing this effect using surface coils. However this approach is often limited by the imaging requirements and is always hard to achieve. Often half volume coils (e.g. open birdcage coils or bent butterfly coils) are used with this type of configuration in order to balance the volume coverage and the geometric decoupling benefits offered by the orthogonal configuration (Adriany and Gruetter 1997). A figure–8 shaped, or butterfly coil, (Figure 4.10b) provides a magnetic field parallel to the circular coil plane, thus significantly reducing the coupling between the two coils. While these elements have shown to be able to produce good geometric decoupling effect, figure–8 coil suffers from low depth penetration thus limiting its uses as a transmit coil (Bottomley et al. 1989). The ideal dual tuned coil that can be designed must have good excitation properties and good efficiencies at two frequencies. Two coil birdcage system offers great benefits over surface coil, such as uniform field excitation, and good geometric decoupling, however birdcage coils are limited due to SNR, and often some more unique anatomy can not be easily imaged using such a configuration. Figure 4.11, shows one possible configuration of a dual tuned volume coil, such design operates at about 90% power efficiency compared to the single tuned coils with the same geometric configuration (Duan et al. 2009). The coupling effects between the two coils are very hard to eliminate especially when they are run in quadrature. 114 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.10: a) Two coils positioned such that the magnetic fields generated by them are orthogonal and thus are naturally decoupled from each other. b) A figure-8 coil (black) or a butterfly coil, generates magnetic field that is perpendicular to the magnetic field generated by circular coil, which is pointing out of the page. Figure 4.11: The layout of a typical dual tuned birdcage coil using crossed elements. Each coil is tuned to a the desired frequency and placed within the other. The runs of the birdcage coil are oriented such that minimal coupling is observed between the two coils. 115 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 4.2.6 Dual Tuned Coil Design for Breast Imaging and Spectroscopy Figure 4.12: Simplified Circuit Diagram of a dual tuned coil. While a number of different techniques can be used to create multiple resonances as describe above, the challenge of accessing the anatomy and intrinsic power limitation of broadband power amplifier guided the design of a dual tuned 31 P/1 H for breast imagind and spectroscopy. A single element surface coil was decided to be best solution to the problem. By integrating the coil within the breast biopsy table, the coil is easily swappable with multichannel 1 H coil array or to perform a biopsy. Dual tuning of a single coil can be analyzed as a LCR circuit and a reactance component made up of an inductor and a capacitor in parallel (Figure 4.12). The total impedance can be written down as the sum of impedances of an LC circuit Z1 plus impedance of an inductor and a capacitor in parallel Z2 . 116 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.13: The RLC circuit has a single resonating frequency ω0 adding a tank circuit to RLC circuit splitts the resonance to produce two resonant frequencies ω1 and ω2 . Z1 = R1 + iXL1 + iXC1 1 1 1 = + Z2 iXL1 iXC1 (4.18) ZT = Z1 + Z2 From equation 4.13 the impedances for individual components can be combined to form: Z = R1 + i ωL1 (1 − L2 1 ) + ω ω 2 L1 C1 1 − ω 2 C2 L2 117 (4.19) PhD Thesis Sergei Obruchkov Dept. of Medical Physics The circuit will have minimal impedance and will resonate when all reactance components are zero this will occur when: C2 = w 2 L1 C1 − 1 + w 2 L2 C1 w2 L2 (w2 L1 C1 − 1) (4.20) Solution to the above equation was performed explicitly which was used to calculate the exact values for the capacitors needed to produced desired resonance frequencies. Value L1 is fixed by the coil geometry, and value of second inductor L2 was chosen to ensure high efficiency on the 31 P channel which is typically L1 /2, then the capacitors values C1 and C2 are calculated for desired frequencies, (See Appendix C ). Efficiency of the dual tuned coil has been shown to largely depend on the values of the inductors used in construction of the coil (Schnall et al. 1985). This is largely due to the fact that the inductors are the most lossy components present in the coil and it is possible to ensure optimal efficiencies for the desired operation using: r E31P = r E1H = L1 L1 + L2 (4.21) L2 L1 + L2 (4.22) 118 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.14: Decoupler hardware from the top: power meter, vector signal generator, 1 kW solid state power amplifier. 4.3 Methods and Materials Several phosphorus coils were designed and built for a GE Signa (GE Medical Systems, Milwaukee, WI) 3T whole body scanner. Broadband capabilities were provided by an accessory consisting of a 4 kW amplifier, T/R switch and a single channel amplifier for 31 P nucleus. The decoupling hardware consisted of a stand alone system with 1 kW power amplifier and is based on Agilent E4438C ESG Vector Signal Generator (Agilent Technologies, Santa Clara, CA) (Figure 4.14). The system was preprogrammed to provide WALTZ– 4 pulse shape modulation, the calibration and control of the which was performed using a PC with an graphical user interface written in Vee Pro (Agilent Technologies, Santa Clara, CA). 119 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.15: RF coil dock, a) TR switch for 31 P, b) 1 H connector, c) cable on the side is the decoupler connection. While the broad band system utilises the same receiver subsystem as the 1 H system, the signal pathway is different, this includes a separate T/R switch and a pre– amplifier. Comparing the broadband pathway to the 1 H pathway is important to insure optimal signal acquisition. This was performed once in order to establish a known initial condition that were later monitored using Quality Assurance QA test designed in house for the broadband system. 4.3.1 31 P Solenoid Coil For System Stability and QA Assurance Tests When designing and building RF coils for multinuclear work, two things have to be considered: a proper transmit receive switch must be used, and unlike 1 H coils transmit and receive coils must be build. Typically coils for multinuclear work are combined i.e. transmit and receive is performed using the same coil. While this usually simplifies the building process, B1 field homogeneity of a coil is limited by its size. 120 PhD Thesis Sergei Obruchkov Dept. of Medical Physics However if the coil is too large the filling factors will decrease resulting in a lowered SNR during receive. Thus transmit–receive coils must balance two parameters: field homogeneity for uniform excitation and good SNR. A 6 turn solenoid coil was tuned and matched for a 1 ml sample of 1M H3 P O4 . The coil was first tuned and matched to the Larmor frequency of 1 H (127.8 MHz at 3.0T). The sample was scanned using a spin echo spectroscopic sequence without water suppression. In order to keep geometry the same the coil was then tuned and matched at the Larmor frequency of 31 P nucleus (51.73 MHz at 3.0 Tesla). The spin echo sequence was used to compare the relative peak height of 1 H and phosphorus, to estimate the performance of the broad band receive chain. This data was later used as a standard for QA tests of 31 P broadband system. Broadband system quality was also verified using a two tone test, by injecting two tones directly into the head–coil carriage coil connectors, and the TR switch using RF signal generator (Model 2030; Marconi Instruments, St. Albans, Colchester). The noise floor was measured and compared using the on board MR receiver system. The noise floor, and peak height was used as a reference for the quality of the 1 H and 31 P receive pathways. Systems docking station shown in figure 4.15 was used in all the experiments. The hydrogen channel of the coil was always connected to either the decouple or the 1 H TR switch. The phosphorus coil was always connected through 31 P TR switch. Adaptors were used to convert the proprietary connections to a female BNC connection. 121 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.16: Simulating double tuned coil using QUCS software. 122 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 4.3.2 Dual Tuned Surface Coil Design and Construction Dual tuned surface coil was designed using the help of simulation software QUCS ( http://qucs.sourceforge.com ). Initial attempt of tuning and matching was performed using optimization algorithm with a search for a global minimum of an S11 parameters at both resonant frequencies. However with the initial design of the coil it was impossible to find values that tuned and matched both channels simultaneously, as the value of matching capacitors on one channel would effect the second channel. A final design of the dual tuned coil shown on figure 4.16, with the blocking circuit followed by the matching network allowed independent matching of the two channels. The tuning and matching was performed using line transformation methods. In practice, parasitics create extra capacitances and inductances throughout the physical circuit which are hard to include into simulations. Thus the simulated values and values used during actual building of the coil differed slightly. By providing the values of the inductors, L1 and L2 the values of the capacitors needed to generate required resonances was calculated using in house software written in Maple (Waterloo Maple Inc., Waterloo, ON). In order to understand the effect of inductors on the circuit, simulation was run multiple times by varying the value of the trap inductor L2 , since L1 is fixed and is determined by the geometry of the coil (Figure 4.17). The results shown in figure 4.18, provide a good estimate of balancing the inductors in order to achieve similar efficiencies on both channels. It was found that using value of L2 = L1 /2 gave the best balanced efficiencies from 123 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.17: Physical layout of the coil the capacitors C1 , C2 and inductors L1 and L2 are responsible for tuning the circuit to the 31 P and 1 H resonant frequencies. the S11 measurements. The final values of L1 and L2 , used to build the coil had efficiencies of 87% and 48% for 31 P and 1 H channel respectively, which is confirmed using equations 4.21 and 4.22. The coil was put into a housing designed to fit into a commercial breast biopsy table, Vanguard ( Sentinelle Medical, Toronto, ON Canada), Figure 4.19. It is possible to swap the dual tuned coil during a clinical breast scan and replace it with the multichannel coil or perform a biopsy. In order to eliminate the chance that the blocking circuits would fail if too much power is fed into one of the ports, the power allowed to enter the coil was limited by software. 124 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.18: Circuit Response while varying the value of the inductor L2 = 100 − −320nH inductor L1 = 300nH is fixed, the tuning and matching capacitors for the circuit are recalculated for each value of L2 to ensure desired center frequencies. Figure 4.19: Coil in the case and in the Sentinelle breast biopsy table. 4.3.3 B1 and B0 Field Calibration Due to low intrinsic concentrations of of 31 P in–vivo, RF power calibration is chal- lenging and unlike 1 H it is not automated on most clinical scanners. Therefore a pulse sequence was written along with reconstruction software in order to calibrate flip angle precisely to 90o . 125 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.20: Transmit gain calibration results window, the ia rf1 value is used to calibrate the power needed for a 90o flip angle. Because it is non trivial to calibrate a slice profile, with a surface transmit coil a 1 ml vial of H3 P O4 was used as a reference for power calibration. Reconstruction software was written to run on the scanner console and present the user with a final power setting value that should be used to produce the desired flip angle (Figure 4.20). The user then sets the value manually and removes the vial from the coil. From the presented user data it is also possible to perform a calculation that will provide a 90o flip angle at a desired depth. This semi automated procedure of transmit power calibration only takes 2–3 minutes and takes away the uncertainty of manual calibration that varies from user to user. Linear and high order shimming (HOS) are used to correct for the inhomogeneities in the magnetic field that the sample introduces in the B0 field when it is placed inside the magnet. Automated shimming can only be performed by collecting 1 H data, and 126 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.21: High order shimming performed using double tuned coil. applying known shim model data, it uses singular value decomposition method to attempt and correct for the introduced inhomogeneities (de Graaf 2007). Shimming the sample decreases the line width of the spectra, thus improving the quality of the collected data. For 31 P spectroscopy, dual tuned coils is a practical option if HOS is required. It is possible to streamline the scanning without moving the patient or change the coils. In house protocol for high order shimming was designed and implemented to use with dual tuned coil. For this purpose 1 H coil port was connected to the scanner (Figure 4.15b) and the 31 P port connected to 31 P T/R switch. The HOS scan was run a several times to ensure that the line width values are minimized (Figure 4.21). 127 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 4.4 Results The dual–tuned coil was first built using air–core high Q inductors (Coilcraft, Cary, IL). After tuning and matching the coil was used, for imaging and spectroscopy work, however, during decoupling tests the inductors have failed. They have been later replaced by in house built inductors made of enamel covered magnet wire. Bench top measurements of the dual tuned coil showed that capacitor and inductor values were very close to the values acquired through the simulations. Since double–tuned coil is a two port device S11 measurements were performed on each port separately. Figure 4.22, shows S11 measurements for 31 P and 1 H channel. The blocking network worked as expected producing good blocking between the two channels. This can be observed from the fact that only a single resonant frequency was present per channel. Dual–tuned coil operate normally with both ports plugged into their respective channels on the scanner and the decoupler. 4.4.1 Coil Performance Imaging, Spectroscopy and Decoupling Images and spectra were successfully acquired using both 31 P and 1 H channels. Be- fore each measurement both ports were plugged into the scanner, the localization and shimming was performed solely using 1 H channel. Then for 31 P channel a semi automated transmit gain calibration and manual centering of the frequency was performed. For the decoupling purposes 1 H port was then disconnected from the scanner and reconnected to the decoupler system. Manual calibration of decoupler power was 128 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.22: Tuning and Matching top Tuning to a desired frequecny, 51.73MHz and 127.8 MHz. Matching to 50 Ω using Smith chart. The blocking circuit performs well, no peaks from the second resonant frequency is observed. performed before 31 P spectra was acquired. The total calibration procedure lasted minimum of 10 minutes. Because this is a transmit receive probe, several interesting challenges with excitation profiles occur. In order to estimate the quality of the coil and the calibration protocol B1 map profile was acquired using 1 H channel. Two different flip angles were used to generate two images from which the B1 map was calculated. It was observed that the flip angle calibration produced proper values as the highest intensity B1 is at the center of the coil (Figure 4.23). Because this is a transmit receive coil, the depth of penetration varies depending on the pulse sequence being used. Gradient and spin 129 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.23: B1 map calculated using two different flip angles. Figure 4.24: Gradient Echo Profile echo sequences with same TE and TR times were used to collect images from a saline phantom. It was observed that the gradient echo sequence produced images with much greater coverage then the spin echo pulse sequence, (Figures 4.24,4.25). The decoupling hardware was first tested for functionality using power attenuator and an oscilloscope. WALTZ–4 sequence with parameters ton = 5ms tof f = 0.5ms, 130 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.25: Spin Echo Profile the nuclear Overhauser effect (NOE) attenuation varied between 0 and 20 dB was used to decouple a 10 cm sphere of methylphosphonic acid (CH3 P (O)(OH)2 ). The center frequency of the decoupler was centered about the hydrogen on the hydroxyl group which resonates as a doublet at 1.2 ppm. The coupling between phosphorus and hydrogen on the hydroxyl group, splits the phosphorous resonance into 4 peaks which are separated by 17.24 Hz (Figure 4.26). After turning on the WALTZ–4 and adjusting the power the peaks collapse into a singlet increasing the SNR 4 times. Due to the chemical structure of methylphosphonic acid no NOE enhancement was observed (Figure 4.26c). No noise injection was observed while running WALTZ–4 decoupling sequence on the proton channel. 4.5 Discussion A dual tuned coil for breast imaging and spectroscopy was successfully designed and built to integrate with Sentinelle’s Vanguard breast biopsy system. This coil allows 131 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 4.26: a) 1 H imaging of polyvinyl alcohol phantom, b) 31 P spectra of methylphosphonic acid c) 31 P spectra of methylphosphonic acid with WALTZ–4 decoupling on 1 H channel. multinuclear data acquisition of 31 P while simultaneously performing decoupling ex- periments on 1 H channel. It is also possible to use this system for further calibration and quantification of 31 P metabolites. Design and implementation of dual tuned RF coil largely depends on the application. For example phosphorus spectroscopy of the breast with decoupling would require a dual tuned coil in order to minimize SAR. However due to limits imposed by the system and the difficulties integrating with the existing hardware, a single dual tuned approach was the simplest to implement. While in general, a departure from single– tuned to dual–tuned coils, will result in loss of performance for at least one of the nuclei, regardless of method used. The 50% loss of sensitivity on hydrogen channel is some what disappointing. This can become an issue from the patient safety point as SAR limit as measured by the system may be reached faster during WALTZ– 4 decoupling. The RF power delivered to the coil is not being transmitted to the patient and instead would be dissipated as heat in the RF components capacitors 132 PhD Thesis Sergei Obruchkov Dept. of Medical Physics and inductors. While there are techniques that can move the coil sensitivities above 90% for both channels their application for breast spectroscopy and integration with clinical work flow environment still needs to be explored (Dabirzadeh et al. 2008; Duan et al. 2009). Multi–nuclear work is slowly becoming popular, particularly as MRI field strengths increase, making clinical spectroscopy possible. This will open new venues for high– performance, versatile multi–frequency probes. In addition, as field strengths of whole–body magnets increase, proton coil designs are becoming increasingly specialized, usually entailing the use of complex multi–channel array coils that cannot easily be re–engineered for multi–frequency use. Therefore many vendors are exploring options for building array of dual tuned coils with parallel transmit and receive capabilities. 133 PhD Thesis Sergei Obruchkov 134 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 5 Design and Construction of an inexpensive Polyvinyl Alcohol Breast Like Phantom for use with MR Imaging This chapter covers construction of a breast–like phantom from polyvinyl alcohol. An inexpensive method of producing PVA cryogels is discussed and the relaxation parameters for 10 and 15 percent concentration of PVA are discussed. 5.1 Introducion Various types of phantoms are used for various purposes in MR imaging and spectroscopy. Phantoms are typically designed to mimic tissue relaxation parameters T1 and T2 or the tissue biochemistry for spectroscopic studies. They also need to mimic the electrical properties of the tissue in order to load the coil properly. Phantoms are typically built to be used during pulse sequence design and optimization, developing MR hardware like RF coils and for quality assurance purposes. MR phantoms are usually designed using water solutions, or various types of hydrogels (e.g. agar, agarose) (Mazzara et al. 1996; Woo et al. 2007; Yoshimura et al. 2003) . While these types of phantoms are easily built and the materials are readily obtainable, they still have some disadvantages. Relaxation parameters T1 and T2 often require addition of relaxation agents such MnCl2 , Aluminum or Gd-DTPA 135 PhD Thesis Sergei Obruchkov Dept. of Medical Physics to change relaxation parameters, which can produce sediment or react and degrade metabolites. Agar gels are susceptible to water loss, and degradation and typically are very fragile. It is also impossible to build the phantoms to have multiple compartments without introducing a water impermeable barrier between different volumes. As well, due to the low separation between agar setting and melting temperatures it is difficult, if not impossible, to introduce layers. Polyvinyl alcohol (PVA), Figure 5.1, is an easily obtainable chemical which is odourless and non-toxic. It is produced from hydrolysis of polyvinyl acetate (commonly used a wood glue). It is known to have an interesting property that by freezing and thawing an aqueous PVA solution promotes cross linking between polymer chains. These type of gels are often called cryogels and have many uses in biotechnology (Lozinsky et al. 2003; Tanabe and Nambu 1988). For example PVA has been used in MRI, Ultrasound (Mano et al. 1986; Chu and Rutt 1997; Surry et al. 2004), CT and X-ray phantoms and also in wound coverings for burn victims. The PVA cryogels can be made mechanically equivalent to tissue (i.e. similar elastic moduli) for ultrasound imaging. T1 and T2 for MR imaging can be tuned by using concentration and by controlling the rate and number of freeze thaw cycles, by increasing the number of cross links via hydrogen bonding in the polymer chains. Lately the PVA cryogels have been extensively studied due to their unique ability to be used in temperature and radiation dosimetry (Zhang and Yu 2004; Lukas et al. 2001). Typically cryogels are hard to produce and require a complicated laboratory. They often require long processing to produce the aqueous solution due to low solubility, 136 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 5.1: Chemical structure of polyvinyl alcohol polymer. and then typically freeze thaw cycles take 12 to 24 hours. However an inexpensive method was developed to produce a breast–like phantom using 15% PVA cryogel, which can be used for MR imaging and spectroscopy and for biopsy training. A vacuum forming system was designed and built to produce a plastic mold to produce phantoms of various types. The use of inexpensive equipment and quick production could be used many times with PVA to make stable phantoms for a variety of imaging experiments. 5.2 Methods and Materials A casting compound (Activia, Marshall, TX) was used to duplicate a mold of a breast from an expensive triple modality breast biopsy phantom (Supertech Elkhart, IN). The cast was then used on a vacuum forming jig, Figure 5.2, to produce a polystyrene negative mold of the breast phantom. 137 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 5.2: Apparatus used for vacuum forming Polystyrene has low glass transition temperature of 95o C and high tensile strength to survive the freeze thaw cycles. A 1.5 mm sheet of polystyrene uniformly heated in an oven and stretched over a cast on the vacuum table was used to produce the desired shape after the styrene had cooled down. Polyvinyl alcohol has low water solubility, and typically reflux condensers are used to boil the PVA solution for two or more hours. Instead a pressure cooker was used as an “autoclave” to simplify the process. A 1L Erlenmeyer flask with the desired concentration of PVA solution was heated in a water bath inside for 30 minutes under 1.2 atmospheres and 125 o C. The flask was weighed before and after to ensure the amount of water remained the same. Once the solution was prepared it needed to be stored in a high humidity environment or used immediately in order to minimize the exposure to air and prevent film formation on the surface. 138 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Aqueous solutions of PVA of 10% and 15% by weight were prepared, and poured into 50 cc plastic vials then placed inside a -20o C freezer for 24 hours alongside a 1 Liter water container. Thawing was performed gradually by keeping the vials in an insulated container alongside frozen water to increase the thermal mass and slow down the thawing process. Vials were first kept in a fridge for 18 hours and then brought to room temperature. Freeze thaw cycles were performed once or twice. Another 15% batch was prepared and poured into the polystyrene breast mold. A 1 cc piece of previously frozen 15% cryogel was embedded into the breast mold before addition of the PVA solution. This was done to mimic a tumour. An inversion recovery pulse sequence was used to calculate T1 value of different samples. In order to speed up the measurements and minimize temperature fluctuations during the scan, the FSE-IR pulse sequence was used. The pulse sequence parameters were TE=16.3 ms, TR=10,000 ms and ETL = 12, inversion time TI varied from 50 to 2500 ms (25 different TI measurement points in total) with a 26cm FOV, slice thickness of 2 cm and acquisition matrix of 256x128. Signal received from the FSE-IR pulse sequence as a function of TI can be written mathematically as: S(T I) = A|1 − 2βe−T I/T1 | (5.1) where A and β are constants. Parameter A ∝ Mo while β denotes the deviation from a perfect 180o pulse. Mean and standard deviation data of the ROI was measured using ImageJ software. Data was fitted using a three parameter least-squares fit 139 PhD Thesis Sergei Obruchkov Dept. of Medical Physics based on the Levenberg-Marquardt (LV) method, to Eq. 5.1. Origin Lab (OriginLab, Northampton, MA) software was used to fit the data using LV algorithm. T2 measurements were done using a spin echo sequence producing 4 echoes from a single excitation. The T2 values were fitted to the spin echo signal equation: S(T E) = Ae −T E T2 . (5.2) 5.3 Results and Discussion The relaxation parameters for the vial samples are shown in Table 5.1. The values compare well with those published in literature (Surry et al. 2004). The T1 and T2 images of 15% PVA breast phantom are shown in Figure 5.3. The simulated tumour had clearly distinguishable contrast. The T2 weighted image shows the contrast inhomogeneity of the phantom. This is probably due to the non-linear spatial freeze thaw temperatures, which introduces realistic fibrous breast contrast. The mock tumour is also easily distinguishable in a T2 weighted image. The PVA phantom is cheap to manufacture and could be used for breast biopsy training, as the mechanical properties of the cryogel are very similar to human tissue. 140 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Table 5.1: T1 and T2 relaxation parameters measured from 10% and 15% PVA samples. Values are mean plus or minus standard deviation, as taken from the ROI placed over the entire 50 cc vial. 10% 15% 1 Freeze Thaw Cycle T1 T2 (ms) (ms) 1571±23 170±33 1450±20 108± 26 10% 15% 2 Freeze Thaw Cycles 1553±17 142±25 1442±12 80±16 PVA concentration Figure 5.3: a) T1 weighted image acquired using 2D FSE-IR, ETL = 6, TR 2200 ms 5 mm slice thickness TE 9.26 ms TI 720 ms, 256x256 b) T2 weighted image acquired using FLAIR , TR 9s TE 140 ms TI 2.25s 256x256 141 PhD Thesis Sergei Obruchkov 142 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 6 Dynamic Measurements of 31P Metabolites in a Posterior Compartment of the Leg During Plantar Flexion Exercise Using Surface Coil or Echo Planar Spectroscopic Imaging for Localization This chapter deals with the use of 31 P MRS for use in dynamic studies that re- quire rapid measurements and spatial localization. In particular a human calf muscle was studied during exercise inside an MRI scanner to dynamically measure relative changes in metabolite concentrations. The EPSI sequence was evaluated for use in dynamic studies. 6.1 Introduction Phosphorus magnetic resonance spectroscopy is often used to access information on the functionality and regulation of the central metabolic pathways of phosphate, i.e. studying the cellular energy capacity. This has been extensively demonstrated to provide a non-invasive in–vivo assessment of muscle metabolism (Carter et al. 2002; Barker and Armstrong 2010). Another key feature of 31 P MRS lies in its ability to provide intracellular pH and ionic magnesium (Mg2+ ) concentration. The pH sensitivity has been extensively used to study muscle during exercise (Morikawa et al. 143 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 1994), brain injury (Rango et al. 1990) and measure intracellular pHi in tumours (Gillies et al. 1994). 31 P MRS is the only technique that can provide non–invasive measurements of pHi . To measure extracellular pHe other techniques such as injection of exogenous compounds like 6-Fluoropyridoxol (19 F MRS) (Mehta et al. 1994), and a 2-imidazole-1-yl-3ethoxycarbonyl propionic acid (IEPA) in 1 H MRS (van Sluis et al. 1999) have been used in animal models. The relative simplicity of a 31 P spectra, is due to the fact that only free or rapidly exchanging metabolites are MR visible, i.e. highly mobile compounds in solution have long transverse relaxation T2 times, (Table 6.1) and appear as sharp peaks. Compounds that are bound to macromolecules or are located in a viscous medium such as the mitochondrial matrix, have short T2 and are essentially MR invisible, i.e. appear as an underlying broad variation of the spectral baseline. The plantar flexion exercise is very simple to perform inside an MRI scanner, requiring movement of the foot towards the shin. It has been shown to be an effective way to exercise the gastrocnemius muscle and can be done to vary the level of work. The bioenergetics pathway involved in energy consumption during muscle contraction is shown in Figure 6.1. During muscle contraction myosine ATPase, utilizes the ATP as energy needed for movement to produce ADP and inorganic phosphate. However the amount of ATP in muscle is only enough to sustain the contractions for a very short period. The back–up energy reserve comes from the phosphocreatine molecule 144 PhD Thesis Sergei Obruchkov Dept. of Medical Physics in the cytoplasm, and creatine kinase catalyzes the reaction to transfer a phosphoryl group from PCr to ADP to produce ATP. So the energy pathway that is typically observed in the NMR time scale can be summarized as; once the energy requirement increases beyond the capacity that can be provided by ATP, the PCr concentration decreases and the amount of inorganic phosphate present in the cytoplasm increases. Concurrent to this process is a decrease in pHi . Under aerobic conditions mitochondrial respiration supplies most of the required energy as ATP, however once the enzyme system is saturated or oxygen is used up only PCr and glycogenolysis can supply ATP. The later produces large amount of lactic acid also decreasing pHi . Dynamic studies demand good temporal resolution while maintaining high enough SNR to achieve useful results. Temporal resolution of a standard Chemical Shift Imaging sequence with a 5x5 CSI grid and TR of 4 seconds, NEX = 1 will only achieve the temporal resolution of 100 seconds. The EPSI with a 5x10 grid and TR of 4 seconds, NEX = 1 can achieve temporal resolution of 20 seconds. The speed up is proportional to the number of phase encoding steps in the x-direction that are encoded using echo planar readout. In contrast, the surface coil itself can be used for localization with temporal resolution being the same as the TR. Size of a receive coil is proportional to its B1 sensitivity profile. The study shows the ability of the EPSI sequence to be useful in dynamic studies while providing spatial information. 145 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.1: Schematic of the main biochemical reactions involved in energy consumption and production during muscle exertion. The rapid depletion of ATP by myosine ATPase in muscle requires a reserve supply provided by phosphocreatine which is used for a rapid ATP production directly in the cytosol of the cell via MM creatine kinase. Table 6.1: Relaxation parameters of 31 P metabolites in a human calf muscle at 25.68 MHz/1.5T and their related chemical shifts (de Graaf 2007, de Certaines 1992). ATP α ATP β ATP γ PCr Pi T1 T2 Chemical Shift (sec) (ms) ppm 3.6 22 -7.52 4.3 8.1 -16.26 -2.48 4.75 16.1 5.517 424.1 0 4.017 204.7 5.02 (pH of 7.2) 6.2 Methods and Materials An in–house built MR compatible ergometer was used for plantar flexion exercises. Three subjects were scanned using the FID CSI or EPSI sequence while performing plantar flexion using 8.3 kg resistance for a period of time followed by a period of rest. The exercise paradigms varied with the experimental protocol (details below). Subjects were placed supine feet first inside the 3T MRI scanner. In order to minimize 146 PhD Thesis Sergei Obruchkov Dept. of Medical Physics motion during the exercise subjects held on to the provided rope handles to counter the force exerted during the plantar flexion exercise. The static 31 P MRS data was acquired using a 5 inch diameter transmit–receive surface coil. The coil was manually tuned and matched to the subject inside the MRI scanner using an MR compatible Network Analyzer, VIA Echo MRI (AEA Technology, Inc. Carlsbad, CA, USA). An EPSI sequence was acquired and compared to a CSI acquisition. A CSI acquisition was acquired using a 5x5 grid with a TR = 4 sec, 16 averages and a FOV = 13 cm, a spectral BW of 5000 Hz and 2048 points, with a soft RF pulse was used for excitation with a slice thickness of 4 cm. The EPSI data set was acquired using a 5x10 grid with a TR = 4 sec, 16 averages and a FOV = 13 cm with a 4 cm slice thickness. The EPSI gradient with a 2.6 cm spatial resolution, spectral resolution of 3125 Hz (1562.5 Hz interleaved twice) and 16,384 points were acquired per interleave. The RF pulse was calibrated using a 4 cc 1x1x4 cm vial of 1 M solution of phosphoric acid and a custom transmit calibration pulse sequence (Section 4.3.3). Data processing of static spectroscopic images was performed using SAGE (GE Healthcarer Milwaukee, WI), while EPSI data was first processed in MATLAB (The Mathworks, Natick, MA, USA) and then exported into SAGE. The same coil was used to acquire dynamic 31 P MRS data. A FID CSI sequence, with no spatial encoding, was used to acquire 16 31 P spectroscopic data points at a temporal resolution of 16 seconds ( acquisition parameters were: 2048 points, spectral 147 PhD Thesis Sergei Obruchkov Dept. of Medical Physics bandwidth 5000 Hz, TR = 8 sec and two averages per data point with π, −π phase cycling). Plantar flexion exercise using 8.3 kg as resistance was performed for a period of 1 minute followed by 3 minutes of rest. The scanner clock was used to guide the subject during the exercise. An EPSI sequence with spatial resolution of 2.6x2.6x4 cm was used to acquire 16 sequential 5x10x820, (5x10 spatial and 820 spectral points), spectroscopic images every 20 seconds. EPSI gradients with spectral bandwidth of 1562.5 Hz, interleaved once, was used to acquire 16,384 data points with TR = 4 sec. For EPSI acquisition the exercise was performed using 8.3 kg resistance for 1 min 40 sec followed by 4 min 40 sec of rest. RF pulse excitation was calibrated using a transmit gain calibration sequence to produce a 90o flip angle at the coil center, where a 1 M vial of phosphoric acid was located. The vial was removed prior to the actual data acquisition. In all cases data was collected after 4 [discarded] excitations in order to achieve steady state prior to data acquisitions. The dynamic data was processed using a combination of Matlab and jMRUI software (http://www.mrui.uab.es/mrui). Peak fitting was performed using using a time-domain AMARES fitting algorithm (Vanhamme et al. 1997) to fit Lorentzian peaks using prior knowledge of peak positions. The pHi calculations were performed using a modified Henderson–Hasselbach relation 1.1 with pKa = 6.77. 148 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.2: CSI and EPSI Data overlaid over the 1 H image of a leg a) 5x5 spectroscopic image of a Leg 13 cm FOV 4cm slice thickness acquired with a TR = 4 seconds, 1 NEX, spectral BW = 5000 Hz, 2048 points. Total scan time 26 minutes 40 seconds. b) 31 P EPSI 5x10 CSI 2.6 cm resolution, spectral BW = 1562.5 Hz interleaved twice, TR = 4 seconds, slice thickness 4 cm. Total scan time 5 minutes 20 seconds. 6.3 Results and Discussion The EPSI sequence was already verified in phantoms, however the short T2 values one would find in vivo are hard to duplicate ex vivo. The acquired spectroscopic image data overlaid over the 1 H image of the leg is shown in Figure 6.2. Overall the 31 P EPSI data shows good correlation with 31 P CSI data. Both FIDs were apodized to 2.4 Hz and the real part of the spectra were manually phased. The SNR was measured from the largest PCr peak in the spectroscopic image. The CSI dataset had SNR of 51.6 and EPSI SNR of 16.8. The PCr FWHM peak widths of both CSI and EPSI were similar 9.4±0.8 Hz. The dynamic data requires high SNR in order to observe peak amplitude and frequency variation. Thus a large voxel is required to produce high SNR when high temporal resolution is needed. Using a surface coil for localization a volume of ap149 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.3: 31 P spectra acquired during plantar flexion exercise using 8.3 kg of resistance. In total 15 spectra were acquired at temporal resolution of 16 seconds. Amplitude variation of PCr and Pi can be clearly observed. proximately 300 cc was acquired using the FID CSI sequence, without any spatial encoding. The real part of the spectra was apodized to 1.8 Hz and manually phased and is shown in Figure 6.3. The depletion of PCr and increase in inorganic phosphate can be clearly seen from the acquired data. The ATP peaks remained unchanged throughout the entire experiment, as expected. The measured SNR of the highest PCr peak was 64.8, and the SNR = 29.2 of Pi was found to be the highest at the end of the exercise, the SNR = 7.8 of ATP was unchanged during the entire dynamic study. The PCr FWHM peak width of the dynamic data was comparable to the CSI and EPSI of 8.12 ± 0.71 Hz. From figure 6.5 the frequency shift of Pi can be clearly observed, this was used to calculate the pHi (Figure 6.6). The dynamic study using EPSI provided similar results but the voxel size of 27 cc was the detrimental factor to the SNR and thus the quality of the spectra. Even after 150 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.4: Dynamic Changes of phosphocreatine and inorganic phosphate peaks during 60 seconds of plantar flexing followed by 3 minutes of rest. apodization of 6.4 Hz the SNR of PCr was only 5.7. Furthermore, the ATP peaks were not resolved and the Pi peak was only observable during the exercise (Figure 6.7). The amplitude of Pi peak was maximal at 100 seconds when the exercise stopped (Figure 6.8). The pHi was calculated using the AMARES algorithm, however due to low SNR peak fitting was suboptimal (Figure 6.9). 151 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.5: 31 P spectra acquired during plantar flexion exercise using 8.3 kg of resistance. Frequency variation of Pi can be clearly observed. Figure 6.6: pHi Calculation of 31 P MRS in an exercising leg without spatial localization, frequency shifts of Pi with respect to PCr show the pH decrease to 6.52 at the end of the exercise. Beginning of recovery was subsequently observed. 152 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.7: Dynamic Study of planar flexion using EPSI 20 second temporal resolution 2.6 cm 5x10 ESPI CSI 1562.5 Hz TR 5 seconds without interleaving. Figure 6.8: Area of Pi and PCr peaks as determined using AMARES algorithm. 153 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Figure 6.9: pH Calculation from frequency shifts of Pi with respect to PCr. 154 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Chapter 7 Conclusions and Future Directions 7.1 Future Applications of the Constructed Double Tuned 1H – flyback 31 31 P Breast RF Coil and the Designed P EPSI Sequence Proton magnetic resonance spectroscopy or hydrogen spectroscopy (1 H-MRS) is the most clinically applied spectroscopy technique today. The success of proton spectroscopy in many parts of the body, however, has been limited due to technical difficulties and limited information content e.g. liver and muscle spectra are relatively simple in comparison to brain spectra due to large dominating lipid resonances. Other NMR visible nuclei such as 31 P, 23 Na, 13 C are either used rarely or almost never for clinical spectroscopy. There are many reasons for this, the first being that most commercial clinical MR systems are designed to work only with 1 H nuclei. Hardware for “multi” nuclear work is often available as an extension to most existing clinical systems however a rough estimate puts less then 5% of all MRI systems in Canada capable of this functionality. The success of proton NMR in clinical applications has only been matched by phosphorus 31 P NMR. A high natural abundance and sensitivity of around 7% that of proton, allows acquisition of a good quality spectrum within clinically relevant time 155 PhD Thesis Sergei Obruchkov Dept. of Medical Physics scales i.e. minutes. A large chemical shift dispersion of 30 ppm, and considerably fewer metabolites than 1 H-NMR, produces “good” peak resolution. Most importantly 31 P spectroscopy is capable of detecting the major compounds responsible for energy metabolism. Furthermore the chemical shift of some phosphorus containing compounds is sensitive to the concentrations of magnesium and pH. The developed rapid chemical shift imaging pulse sequences, 31 P Echo Planar Spec- troscopic Imaging, has been shown to be robust in acquiring 2D spectral images using a 3T Signa Excite MR Scaner (GE Healthcare, Milwaukee IL USA) located at St Joseph’s Healthcare in Hamilton Ontario, which is equipped with multi nuclear capabilities. Combining the 31 P EPSI sequence with the in–house developed auto prescan func- tionality for multi nuclear spectroscopy enables the calibration to be done semi automatically, drastically reducing setup and prescan time while removing the operator dependent inconsistencies. Furthermore, the ability to use dual tuned coil for higher order shimming and decoupling minimizes the setup time, moving patients in order to swap coils or alter the setup. The project was oriented towards technical development of software and hardware with the end goal to utilize this approach in a clinical population. The future work will involve using the EPSI technique for studying breast cancer and muscle metabolism. Using an MRI compatible breast imaging and biopsy system (Sentinelle Medical, Toronto Ontario Canada), available at St. Joseph’s Healthcare to couple rapid spec156 PhD Thesis Sergei Obruchkov Dept. of Medical Physics troscopy with breast biopsy is ideal for studying breast cancer. Current MRI literature of 31 P spectroscopy presents clear evidence of increased concentrations of phospho- monoesters (PMEs), and phosphodiesters (PDEs) as well as inorganic phosphate in breast tumor tissue (Leach et al. 1998; Arias-Mendoza et al. 2004). Acquisition of 31 P spectra is complex due to lack of developed hardware and software for clinical scanners, this is further hampered with complex protocol development required to complete clinical scans. However this thesis has successfully showed that it is possible to acquire 31 P MRS in as little as 20 seconds with voxel size larger then 40 cc, and use this information for dynamic studies. Furthermore, the advantages of 31 P EPSI sequence were observed in acquisition with high signal especially when large number of voxel counts are required. The acquisition speed increase of up to 20 times could be achieved provided that the SNR can be sacrificed. The fundamental limit of SNR remains the magnetic field strength and the optimization of the coils. Current state of the art MR systems all utilize volume transmit and surface coil arrays to receive the signal. The smaller surface arrays drastically increased the SNR for 1 H imaging purposes, by coupling less noise from the sample. The new surface coil array technology utilises low noise amplifiers with noise figures as low as 0.35 that can boost SNR by more than 40%. Utilizing the phased array coil technology for 31 P MRS is challenging but not impossible, the difficulty lies in the fact that the bore of the MR machine has limited size and must contain body coil tuned to transmit only at 1 H frequencies, shim coils, gradient coils, insulation and cooling. This drastically hampers the development of new devices optimized for 157 PhD Thesis Sergei Obruchkov Dept. of Medical Physics other nuclei. One solution would be to have a large dual tuned transmit coil and design surface receive arrays tuned for either 31 P or 1 H frequencies. Another solution may come in the development of compact transmit/receive arrays that could provide more B1 field homogeneity. The dual tuned surface coil in this thesis has advantages that it remains compact and mobile and it is easy to swap in and out with the existing 1 H hardware, while still allowing for complex experiments require transmit at multiple frequencies. This technology could easily be extended in designing small compact arrays for improving spatial coverage and increasing SNR. Using technology developed in this thesis a future study will be designed to collect phosphorus spectra from the various clinical populations. First the protocols will be designed and verified on phantoms. Subsequently clinical populations with breast cancer, liver diseases and metabolic muscular diseases could be scanned using these developed techniques. The studies could be easily designed such that they will be performed along side of clinically diagnostic procedures during which the biopsy will be performed by trained medical staff. The results from spectroscopy and biopsy can be used to develop new non–invasive protocols with high sensitivity and specificity. 158 PhD Thesis Sergei Obruchkov Dept. of Medical Physics 7.2 Future Directions and other Applications of ESPI sequence and Multi Tuned MR Coils. Using RF coils to localize the signal is advantageous in dynamic studies, while EPSI is better suited for speeding up high resolution CSI acquisition as shown in Chapter 3. Current CSI acceleration techniques have their challenges. It has been shown that EPSI with a flyback trajectory is a strong contender enabling researchers to speed up the acquisition process (Zierhut et al. 2009). However experimental design, SNR requirements and artifact severity have to be considered. The biggest challenge currently present for use of the 31 P EPSI sequence is the poor SNR due to poor sensitivities and low metabolic concentrations. One key approach to improving multinuclear SNR is to use phased array coils. In particular the design of double tuned multinuclear coils is of current investigational interest (Lee et al. 2000; Avdievich and Hetherington 2009; Avdievich and Hetherington 2007). The phased array multinuclear coils require uniform B1 field for excitation, which presents an extra layer of complication. Another point to consider is that as individual elements of a phased array coil become minimized the coil thermal noise begins to dominate over patient noise. At this point any further SNR improvement would require use of cryogens to reduce the coil Johnson noise thereby increasing SNR (Ginefri et al. 2007; Darrasse and Ginefri 2003). To allow for the CSI spatial encoding gradients 0.2 ms is required after the RF excitation. This is potentially a long time, especially when trying to resolve species with 159 PhD Thesis Sergei Obruchkov Dept. of Medical Physics short T2 s, such as31 P and 23 Na. Alternative techniques could be used to acquire data immediately following excitation thereby maximizing SNR of these species (Robson et al. 2005). 31 P MRS requires a larger receiver bandwidth then 1 H MRS, demanding higher per- formance from the gradients (i.e. faster slew rates and higher amplitudes). While interleaving gradients is a robust option to eliminate the need for better gradients, other techniques of sub sampling MRS data during dynamic acquisitions and using prior knowledge for data reconstruction (e.g. a coil sensitivity profile) are also possible to speed up the acquisition (Otazo et al. 2007; Lin et al. 2007). For clinical work, 31 P EPSI is well suited as it requires minimal work flow modifi- cations and can be utilized with commercially available hardware. For breast spectroscopic imaging (i.e. non-dynamic needs) the current setup is ready for a clinical trial. To boost SNR over the slice of acquisition more averaging will be required while EPSI will still keep the acquisition well within a clinically feasible time frame (30-40 minutes). The biggest future challenges will be in dynamic studies, where speed is essential. 160 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Bibliography Adriany, G. and R. Gruetter (1997, Mar). A half-volume coil for efficient proton decoupling in humans at 4 tesla. J Magn Reson 125 (1), 178–184. Arias-Mendoza, F., K. Zakian, A. Schwartz, F. A. Howe, J. A. Koutcher, M. O. Leach, J. R. Griffiths, A. Heerschap, J. D. Glickson, S. J. Nelson, J. L. Evelhoch, H. C. Charles, T. R. Brown, and C. G. on MRS Applications in Cancer (2004, Oct). Methodological standardization for a multi-institutional in vivo trial of localized 31p mr spectroscopy in human cancer research. in vitro and normal volunteer studies. NMR Biomed 17 (6), 382–391. Avdievich, N. I. and H. P. Hetherington (2007, Jun). 4 t actively detuneable doubletuned 1h/31p head volume coil and four-channel 31p phased array for human brain spectroscopy. J Magn Reson 186 (2), 341–346. Avdievich, N. I. and H. P. Hetherington (2009, Nov). High-field head radiofrequency volume coils using transverse electromagnetic (tem) and phased array technologies. NMR Biomed 22 (9), 960–974. Barker, A. R. and N. Armstrong (2010, Aug). Insights into developmental muscle metabolism through the use of 31p-magnetic resonance spectroscopy: a review. Pediatr Exerc Sci 22 (3), 350–368. Barker, P. B., X. Golay, D. Artemov, R. Ouwerkerk, M. A. Smith, and A. J. Shaka (2001, Feb). Broadband proton decoupling for in vivo brain spectroscopy in humans. Magn Reson Med 45 (2), 226–232. Beloueche-Babari, M., Y.-L. Chung, N. M. S. Al-Saffar, M. Falck-Miniotis, and M. O. Leach (2010, Jan). Metabolic assessment of the action of targeted cancer therapeutics using magnetic resonance spectroscopy. Br J Cancer 102 (1), 1–7. Bomsdorf, H., P. Röschmann, and J. Wieland (1991, Nov). Sensitivity enhancement in whole-body natural abundance 13c spectroscopy using 13c/1h doubleresonance techniques at 4 tesla. Magn Reson Med 22 (1), 10–22. Bottomley, P. A. and C. J. Hardy (1989, Mar). Rapid, reliable in vivo assays of human phosphate metabolites by nuclear magnetic resonance. Clin Chem 35 (3), 392–395. Bottomley, P. A., C. J. Hardy, P. B. Roemer, and O. M. Mueller (1989, Dec). Proton-decoupled, overhauser-enhanced, spatially localized carbon-13 spectroscopy in humans. Magn Reson Med 12 (3), 348–363. Brown, T., K. Ugurbil, and R. Shulman (1977). 31p nuclear magnetic resonance measurements of atpase kinetics in aerobic escherichia coli cells. Proc Natl Acad Sci 74, 5551–5553. 161 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Carter, G. T., R. T. Abresch, and W. M. Fowler (2002, Nov). Adaptations to exercise training and contraction-induced muscle injury in animal models of muscular dystrophy. Am J Phys Med Rehabil 81 (11 Suppl), S151–S161. Chen, A. P., C. H. Cunningham, E. Ozturk-Isik, D. Xu, R. E. Hurd, D. A. C. Kelley, J. M. Pauly, J. Kurhanewicz, S. J. Nelson, and D. B. Vigneron (2007, Jun). High-speed 3t mr spectroscopic imaging of prostate with flyback echoplanar encoding. J Magn Reson Imaging 25 (6), 1288–1292. Chen, C., D. Hoult, and V. Sank (1983). Quadrature detection coils a further 21/2 improvement in sensitivity. Journal of magnetic resonance 54 (2), 324–327. Chu, K. C. and B. K. Rutt (1997, Feb). Polyvinyl alcohol cryogel: an ideal phantom material for mr studies of arterial flow and elasticity. Magn Reson Med 37 (2), 314–319. Cunningham, C. H., D. B. Vigneron, A. P. Chen, D. Xu, S. J. Nelson, R. E. Hurd, D. A. Kelley, and J. M. Pauly (2005, Nov). Design of flyback echo-planar readout gradients for magnetic resonance spectroscopic imaging. Magn Reson Med 54 (5), 1286–1289. Dabirzadeh, A., C.-W. Chang, and M. P. McDougall (2008). An insertable p-31 rf coil for dual-frequency magnetic resonance imaging & spectroscopy. Conf Proc IEEE Eng Med Biol Soc 2008, 2036–2038. Darrasse, L. and J.-C. Ginefri (2003, Sep). Perspectives with cryogenic rf probes in biomedical mri. Biochimie 85 (9), 915–937. de Certaines, J., W. Bovee, and F. Podo (1992). Magnetic Resonance Spectroscopy in Biology and Medicine. Pergamon Press. de Graaf, R. (2005a). Metabolomics by In Vivo NMR, Chapter In Vivo NMR spectroscopy, pp. 7–21. Wiley. Chapter by de Graaf, R. de Graaf, R. (2007). In vivo NMR spectroscopy. Principles and Techniques (2 ed.). Wiley. de Graaf, R. A. (1998). In vivo NMR spectroscopy. Principles and techniques. John Wiley. de Graaf, R. A. (2005b, Jun). Theoretical and experimental evaluation of broadband decoupling techniques for in vivo nuclear magnetic resonance spectroscopy. Magn Reson Med 53 (6), 1297–1306. de Graaf, R. A., K. Nicolay, and M. Garwood (1996, May). Single-shot, b1insensitive slice selection with a gradient-modulated adiabatic pulse, biss-8. Magn Reson Med 35 (5), 652–657. Dietrich, O., M. F. Reiser, and S. O. Schoenberg (2008, Jan). Artifacts in 3-t mri: physical background and reduction strategies. Eur J Radiol 65 (1), 29–35. 162 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Duan, Y., B. S. Peterson, F. Liu, T. R. Brown, T. S. Ibrahim, and A. Kangarlu (2009, Jan). Computational and experimental optimization of a double-tuned (1)h/(31)p four-ring birdcage head coil for mrs at 3t. J Magn Reson Imaging 29 (1), 13–22. Dydak, U. and M. Schar (2006, May). Mr spectroscopy and spectroscopic imaging: comparing 3.0 t versus 1.5 t. Neuroimaging Clin N Am 16 (2), 269–83, x. Evelhoch, J., M. Crowley, and J. Ackerman (1984). Signal-to-noise optimization and observed volume localization with circular surface coils. Journal of Magnetic Resonance (1969) 56 (1), 110–124. Fitzsimmons, J. R., B. L. Beck, and H. R. Brooker (1993, Jul). Double resonant quadrature birdcage. Magn Reson Med 30 (1), 107–114. Fitzsimmons, J. R., H. R. Brooker, and B. Beck (1987, Nov). A transformercoupled double-resonant probe for nmr imaging and spectroscopy. Magn Reson Med 5 (5), 471–477. Forsen, S. and R. Hoffman (1963). Study of moderately rapid chemical exchange reactions by means of nuclear magnetic double resonance. J Chem Phys 39, 2892–2901. Garwood, M. and L. DelaBarre (2001, Dec). The return of the frequency sweep: designing adiabatic pulses for contemporary nmr. J Magn Reson 153 (2), 155– 177. Gillies, R. J., Z. Liu, and Z. Bhujwalla (1994, Jul). 31p-mrs measurements of extracellular ph of tumors using 3-aminopropylphosphonate. Am J Physiol 267 (1 Pt 1), C195–C203. Gillies, R. J. and D. L. Morse (2005). In vivo magnetic resonance spectroscopy in cancer. Annu Rev Biomed Eng 7, 287–326. Ginefri, J.-C., M. Poirier-Quinot, O. Girard, and L. Darrasse (2007, Sep). Technical aspects: development, manufacture and installation of a cryo-cooled hts coil system for high-resolution in-vivo imaging of the mouse at 1.5 t. Methods 43 (1), 54–67. Gonen, O., S. Gruber, B. S. Li, V. Mlynárik, and E. Moser (2001, Oct). Multivoxel 3d proton spectroscopy in the brain at 1.5 versus 3.0 t: signal-to-noise ratio and resolution comparison. AJNR Am J Neuroradiol 22 (9), 1727–1731. Gorter, C. (1936). Negative results of an attempt to detect nuclear magnetic spins. Physica 3, 503. Griffiths, D. and C. Inglefield (1999). Introduction to electrodynamics. Prentice Hall New Jersey;. Harrington, R. (1993). Field computation by moment methods. Wiley-IEEE Press. Hayes, C. E. (2009, Nov). The development of the birdcage resonator: a historical perspective. NMR Biomed 22 (9), 908–918. 163 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Hetherington, H. P., J. W. Pan, W. J. Chu, G. F. Mason, and B. R. Newcomer (1997, Dec). Biological and clinical mrs at ultra-high field. NMR Biomed 10 (8), 360–371. Hoult, D. I., S. J. Busby, D. G. Gadian, G. K. Radda, R. E. Richards, and P. J. Seeley (1974, Nov). Observation of tissue metabolites using 31p nuclear magnetic resonance. Nature 252 (5481), 285–287. Jesmanowicz, A., S. Li, and J. S. Hyde (2002). Simultaneous acquisition of echo planar images on proton and phosphor frequencies. In Proceedings of the ISMRM, pp. 2501. Jung, Y., Y. Jashnani, R. Kijowski, and W. F. Block (2007, Jan). Consistent non-cartesian off-axis mri quality: calibrating and removing multiple sources of demodulation phase errors. Magn Reson Med 57 (1), 206–212. Keevil, S. F. (2006, Aug). Spatial localization in nuclear magnetic resonance spectroscopy. Phys Med Biol 51 (16), R579–R636. Kim, H., R. B. Thompson, C. C. Hanstock, and P. S. Allen (2005, Apr). Variability of metabolite yield using steam or press sequences in vivo at 3.0 t, illustrated with myo-inositol. Magn Reson Med 53 (4), 760–769. Krüger, G., A. Kastrup, and G. H. Glover (2001, Apr). Neuroimaging at 1.5 t and 3.0 t: comparison of oxygenation-sensitive magnetic resonance imaging. Magn Reson Med 45 (4), 595–604. Kurhanewicz, J., D. B. Vigneron, and S. J. Nelson (2000). Three-dimensional magnetic resonance spectroscopic imaging of brain and prostate cancer. Neoplasia 2 (1-2), 166–189. Leach, M. O., M. Verrill, J. Glaholm, T. A. Smith, D. J. Collins, G. S. Payne, J. C. Sharp, S. M. Ronen, V. R. McCready, T. J. Powles, and I. E. Smith (1998, Nov). Measurements of human breast cancer using magnetic resonance spectroscopy: a review of clinical measurements and a report of localized 31p measurements of response to treatment. NMR Biomed 11 (7), 314–340. Lee, R. F., R. Giaquinto, C. Constantinides, S. Souza, R. G. Weiss, and P. A. Bottomley (2000, Feb). A broadband phased-array system for direct phosphorus and sodium metabolic mri on a clinical scanner. Magn Reson Med 43 (2), 269– 277. Leroy-Willig, A., Y. Fromes, M. Paturneau-Jouas, and P. Carlier (2003). Assessing gene and cell therapies applied in striated skeletal and cardiac muscle: Is there a role for nuclear magnetic resonance? Neuromuscular Disorders 13 (5), 397–407. Lin, F.-H., S.-Y. Tsai, R. Otazo, A. Caprihan, L. L. Wald, J. W. Belliveau, and S. Posse (2007, Feb). Sensitivity-encoded (sense) proton echo-planar spectroscopic imaging (pepsi) in the human brain. Magn Reson Med 57 (2), 249–257. 164 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Lozinsky, V. I., I. Y. Galaev, F. M. Plieva, I. N. Savina, H. Jungvid, and B. Mattiasson (2003, Oct). Polymeric cryogels as promising materials of biotechnological interest. Trends Biotechnol 21 (10), 445–451. Lukas, L. A., K. J. Surry, and T. M. Peters (2001, Nov). Temperature dosimetry using mr relaxation characteristics of poly(vinyl alcohol) cryogel (pva-c). Magn Reson Med 46 (5), 1006–1013. Mano, I., H. Goshima, M. Nambu, and M. Iio (1986, Dec). New polyvinyl alcohol gel material for mri phantoms. Magn Reson Med 3 (6), 921–926. Matson, G. B., P. Vermathen, and T. C. Hill (1999, Jul). A practical double-tuned 1h/31p quadrature birdcage headcoil optimized for 31p operation. Magn Reson Med 42 (1), 173–182. Mazzara, G. P., R. W. Briggs, Z. Wu, and B. G. Steinbach (1996). Use of a modified polysaccharide gel in developing a realistic breast phantom for mri. Magn Reson Imaging 14 (6), 639–648. Mehta, V. D., P. V. Kulkarni, R. P. Mason, A. Constantinescu, S. Aravind, N. Goomer, and P. P. Antich (1994, Aug). 6-fluoropyridoxol: a novel probe of cellular ph using 19f nmr spectroscopy. FEBS Lett 349 (2), 234–238. Merkle, E. M. and B. M. Dale (2006, Jun). Abdominal mri at 3.0 t: the basics revisited. AJR Am J Roentgenol 186 (6), 1524–1532. Mierisova, S. and M. Ala-Korpela (2001, Jun). Mr spectroscopy quantitation: a review of frequency domain methods. NMR Biomed 14 (4), 247–259. Moon, R. B. and J. H. Richards (1973, Oct). Determination of intracellular ph by 31p magnetic resonance. J Biol Chem 248 (20), 7276–7278. Morikawa, S., T. Inubushi, K. Kito, and R. Tabata (1994). Imaging of phosphoenergetic state and intracellular ph in human calf muscles after exercise by 31p nmr spectroscopy. Magn Reson Imaging 12 (7), 1121–1126. Murphy-Boesch, J., H. Jiang, R. Stoyanova, and T. R. Brown (1998, Mar). Quantification of phosphorus metabolites from chemical shift imaging spectra with corrections for point spread effects and b1 inhomogeneity. Magn Reson Med 39 (3), 429–438. Murphy-Boesch, J., R. Srinivasan, L. Carvajal, and T. R. Brown (1994, Feb). Two configurations of the four-ring birdcage coil for 1h imaging and 1h-decoupled 31p spectroscopy of the human head. J Magn Reson B 103 (2), 103–114. Oppenheim, A., R. Schafer, and J. Buck (1998). Discrete Time Signal Processing (2nd ed.). Prentice Hall. Otazo, R., S.-Y. Tsai, F.-H. Lin, and S. Posse (2007, Dec). Accelerated short-te 3d proton echo-planar spectroscopic imaging using 2d-sense with a 32-channel array coil. Magn Reson Med 58 (6), 1107–1116. 165 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Pan, J. W., J. R. Hamm, D. L. Rothman, and R. G. Shulman (1988, Nov). Intracellular ph in human skeletal muscle by 1h nmr. Proc Natl Acad Sci U S A 85 (21), 7836–7839. Pohmann, R., M. von Kienlin, and A. Haase (1997, Dec). Theoretical evaluation and comparison of fast chemical shift imaging methods. J Magn Reson 129 (2), 145–160. Posse, S., C. DeCarli, and D. L. Bihan (1994, Sep). Three-dimensional echo-planar MR spectroscopic imaging at short echo times in the human brain. Radiology 192 (3), 733–738. Provencher, S. W. (1993, Dec). Estimation of metabolite concentrations from localized in vivo proton nmr spectra. Magn Reson Med 30 (6), 672–679. Rango, M., R. E. Lenkinski, W. M. Alves, and T. A. Gennarelli (1990, Nov). Brain ph in head injury: an image-guided 31p magnetic resonance spectroscopy study. Ann Neurol 28 (5), 661–667. Robson, M. D., J. C. Gore, and R. T. Constable (1997, Nov). Measurement of the point spread function in mri using constant time imaging. Magn Reson Med 38 (5), 733–740. Robson, M. D., D. J. Tyler, and S. Neubauer (2005, Feb). Ultrashort te chemical shift imaging (ute-csi). Magn Reson Med 53 (2), 267–274. Ronen, S. M., E. Rushkin, and H. Degani (1992, Mar). Lipid metabolism in large t47d human breast cancer spheroids: 31p- and 13c-nmr studies of choline and ethanolamine uptake. Biochim Biophys Acta 1138 (3), 203–212. Sacolick, L. I., D. L. Rothman, and R. A. de Graaf (2007, Mar). Adiabatic refocusing pulses for volume selection in magnetic resonance spectroscopic imaging. Magn Reson Med 57 (3), 548–553. Schnall, M., V. Harihara Subramanian, J. Leigh, et al. (1985). A new double-tuned probed for concurrent 1H and 31P NMR* 1. Journal of Magnetic Resonance (1969) 65 (1), 122–129. Slichter, C. (1991). Principles of Magnetic Resonance. Springer. Staewen, R. S., A. J. Johnson, B. D. Ross, T. Parrish, H. Merkle, and M. Garwood (1990, May). 3-d flash imaging using a single surface coil and a new adiabatic pulse, bir-4. Invest Radiol 25 (5), 559–567. Surry, K. J. M., H. J. B. Austin, A. Fenster, and T. M. Peters (2004, Dec). Poly(vinyl alcohol) cryogel phantoms for use in ultrasound and mr imaging. Phys Med Biol 49 (24), 5529–5546. Tanabe, T. and M. Nambu (1988, March 29). Medical material of polyvinyl alcohol and process of making. US Patent 4,734,097. 166 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Ulrich, M., T. Wokrina, G. Ende, M. Lang, and P. Bachert (2007, Apr). 31p-1H echo-planar spectroscopic imaging of the human brain in vivo. Magn Reson Med 57 (4), 784–790. van Sluis, R., Z. M. Bhujwalla, N. Raghunand, P. Ballesteros, J. Alvarez, S. Cerdán, J. P. Galons, and R. J. Gillies (1999, Apr). In vivo imaging of extracellular ph using 1h mrsi. Magn Reson Med 41 (4), 743–750. Vanhamme, van den Boogaart A, and V. H. S (1997, Nov). Improved method for accurate and efficient quantification of mrs data with use of prior knowledge. J Magn Reson 129 (1), 35–43. Vanhamme, L., T. Sundin, P. V. Hecke, and S. V. Huffel (2001, Jun). Mr spectroscopy quantitation: a review of time-domain methods. NMR Biomed 14 (4), 233–246. Walter, G., E. R. Barton, and H. L. Sweeney (2000, May). Noninvasive measurement of gene expression in skeletal muscle. Proc Natl Acad Sci U S A 97 (10), 5151–5155. Weyers, D. (2009, 08). Private discussions. Discussing MR coils with multiple resonances. Widmaier, S., J. Breuer, W. I. Jung, G. J. Dietze, and O. Lutz (1998, Sep). 31p/1h waltz-4 broadband decoupling at 1.5 t: different versions of the composite pulse and consequences when using a surface coil. Magn Reson Imaging 16 (7), 845– 849. Woo, D.-C., B.-S. Kim, S.-L. Jung, H.-J. Park, H.-S. Rhim, G.-H. Jahng, and B.-Y. Choe (2007, May). Development of a cone-shape phantom for multi-voxel mr spectroscopy. J Neurosci Methods 162 (1-2), 101–107. Yoshimura, K., H. Kato, M. Kuroda, A. Yoshida, K. Hanamoto, A. Tanaka, M. Tsunoda, S. Kanazawa, K. Shibuya, S. Kawasaki, and Y. Hiraki (2003, Nov). Development of a tissue-equivalent mri phantom using carrageenan gel. Magn Reson Med 50 (5), 1011–1017. Zhang, S.-J. and H.-Q. Yu (2004, Jan). Radiation-induced degradation of polyvinyl alcohol in aqueous solutions. Water Res 38 (2), 309–316. Zhu, X., A. Ebel, J. X. Ji, and N. Schuff (2007, May). Spectral phase-corrected grappa reconstruction of three-dimensional echo-planar spectroscopic imaging (3d-epsi). Magn Reson Med 57 (5), 815–820. Zierhut, M. L., E. Ozturk-Isik, A. P. Chen, I. Park, D. B. Vigneron, and S. J. Nelson (2009, Sep). (1)h spectroscopic imaging of human brain at 3 tesla: comparison of fast three-dimensional magnetic resonance spectroscopic imaging techniques. J Magn Reson Imaging 30 (3), 473–480. 167 PhD Thesis Sergei Obruchkov 168 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Appendix A function [ epsigrad , wave_form , my_hdr ] = epsigrad ( BW , RES , Sr ) % Generate EPSI g r a d i e n t s by g i v i n g p a r a m e t e r s % % I n p u t : Bandwidth R e s o l u t i o n , and S l e w r a t e % BW i n Hz % RES i n cm % Sr i n Gaus/cm/ s % % Output : % epsigrad . half H a l f t r a p e z o i d s h o r t waveform % . full F u l l waveform up t o max p u l s e w i d t h up t o 16384 p o i n t s % . c o m p l e t e two waveform s t i t c h e d t o g e t h e r % wave form Exact waveform % my hdr Header f o r t h e c r e a t e d g r a d i e n t s % % EPIC NOTE: MAKES SURE THAT THE FILES ARE EVEN XLATEBIN % DOES NOT PRODUCE A PROPER HEADER/CHECKSUM OTHERWISE ! ! ! % % Example : % [ e p s i , wave , hdr ]= e p s i g r a d ( 2 0 0 0 , 2 , 1 5 0 0 0 ) ; % % Test t h e waves u s i n g e p s i s a m p l i n g % Rewritten % Jan 15 t h 2009 , Matlab 7 . 6 R. 2 0 0 8 a % % Mar 5 t h 2 0 0 9 : % A f t e r g r a d t e s t s found o u t t h a t a r e a i s s t i l l o n l y h a l f % o f what i t s u p p o s e d t o be . M u l t i p l i e d a l l a r e a c a l c s by 2 . % S e r g e i Obruchkov %% Args Test i f nargin == 0 disp ( ' See ” h e l p e p s i g r a d ” f o r arguments ' ) end i f nargin < 1 BW = input ( ' S p e c t r a l BandWidth [ 2 0 0 0 ] ( Hz ) : ' ) ; i f isempty ( BW ) BW = 2 0 0 0 ; end end i f nargin < 2 RES = input ( ' S p a t i a l R e s o l u t i o n [ 2 . 0 ] (cm) : ' ) ; i f isempty ( RES ) RES = 2 . 0 ; end end i f nargin < 3 Sr = input ( 'Max SlewRate o f g r a d i e n t s [ 1 5 0 0 0 ] ( Gauss /cm/ s ) : ' ) ; i f isempty ( Sr ) Sr = 1 5 0 0 0 . 0 ; end end %% MAIN %G l o b a l v a r i a b l e s global gamma; %gamma = 4 2 . 5 7 5 e3 ; gamma = 1 7 . 2 3 5 e3 ; area = calcArea ( BW , RES , Sr ) ; 169 PhD Thesis Sergei Obruchkov Dept. of Medical Physics [ hrewind , trap_rewind , my_hdr ] = calcParams ( BW , RES , Sr , area ) ; my_hdr . Area_req = 1 0 / (gamma∗ RES ) ; % Check t h a t a r e a s A r e q r e q matches Area k wave_form = cat ( 1 , hrewind , trap_rewind ) ; % I n t e r p o l a t i n g waveforms t o 4 us r e s o l u t i o n hrewind_d = wavetoGradUP ( hrewind ) ; trap_rewind_d = wavetoGradUP ( trap_rewind ) ; half = convtoEvenINT ( hrewind_d ) ; f u l l = convtoEvenINT ( trap_rewind_d ) ; % c h e c k and c o r r e c t t h a t a r e a s match i f sum( f u l l ) ˜= 0 [ C , I ] = max( f u l l ) ; f u l l ( I −1) = f u l l ( I −1) − sum( f u l l ) ; end area_diff = sum( abs ( f u l l ) ) − sum( abs ( half ) ) ∗ 4 ; i f area_diff ˜= 0 [ C , I ] = min( half ) ; half ( I ) = half ( I ) − area_diff /4 + mod ( area_diff / 4 , 2 ) ; end % check t h a t f u l l or h a l f do n o t have two z e r o e s a t t h e end Lful = length ( f u l l ) ; Lhalf = length ( half ) ; i f f u l l ( Lful −1) == 0 f u l l ( Lful ) = [ ] ; end i f half ( Lhalf −1) == 0 half ( Lhalf ) = [ ] ; end Lful = length ( f u l l ) ; %Get new l e n g t h % F i n a l i z e t h e waveforms % h a l f i s a s h o r t waveform and d o e s n o t r e p e a t % f u l l w i l l r e p e a t s u g e s t e d NumberofLobes t o max o u t t h e time p u l s e w i d t h o f 65536 ←us % or 16384 p o i n t s w i t h 4 us time r e s epsigrad . half=cat ( 2 , half , 1 ) ; epsigrad . f u l l = cat ( 2 , f u l l , 1 ) ; % F i n a l i z i n g header my_hdr . half . Resolution = length ( epsigrad . half ) −1; my_hdr . half . PulseWidth = ( length ( epsigrad . half ) −1) ∗ 4 ; my_hdr . f u l l . Resolution = length ( epsigrad . f u l l ) −1; my_hdr . f u l l . PulseWidth = ( length ( epsigrad . f u l l ) −1) ∗ 4 ; my_hdr . SugestedminTR= ( 1 0 2 4 ∗ ( length ( epsigrad . f u l l ) −1)∗4+( length ( epsigrad . half ) −1)←∗ 4 ) / 1 0 0 0 ; % minTR t o u s e i n c v e v a l i n ms % For s i m u l a t o r % 1024 l o b e s simulate = cat ( 2 , half , f u l l ( 2 : Lful ) ) ; for n = 1:1024 simulate = cat ( 2 , simulate , f u l l ( 2 : Lful ) ) ; end epsigrad . simulate = cat ( 1 , 1 : 4 : 4 ∗ length ( simulate ) , double ( simulate ) ) ;% time i n us and←amplitude 170 PhD Thesis Sergei Obruchkov Dept. of Medical Physics % MAIN END % C a l c u l a t e needed p a r a m e t e r s function Area_r = calcArea ( BW , RES , Sr ) ; % Global Constants : % Gamma o v e r 2∗ Pi f o r 31P i n Hz/ Gauss = 1 7 . 2 3 5 e3 % Gamma o v e r 2∗ Pi f o r 1H i n Hz/ Gauss = 4 2 . 5 7 5 e3 %gamma = 4 2 . 5 7 5 e3 ; %gamma = 1 7 . 2 3 5 e3 ; % Area % i .e. % AREA global needed t o f l y t h r o u g h t h e needed r e s o l u t i o n Area needed t o t r a v e l k−s p a c e = x r e s c s i ∗10/(2∗gamma∗FOV) gamma; area =10/(gamma∗ RES ) ; eq1 eq2 eq3 eq4 = = = = [ [ [ [ ' ( 1 / ( 2 ∗BW)−a r e a / ( 2 ∗ Sr ∗ r ) − r ) ˆ2 − a r e a / Sr − r ˆ2 ' ] ; 'BW = ' , num2str ( BW ) ] ; ' a r e a = ' , num2str ( area ) ] ; ' Sr = ' , num2str ( Sr ) ] ; soln = solve ( eq1 , eq2 , eq3 , eq4 ) ; %P i c k i n g t h e a p r o p r i a t e s o l u t i o n % I f t h e r e are i f i s r e a l ( soln . r ) r = eval ( soln . r ( 2 ) ) ; else r = eval ( soln . r ( 1 ) ) ; end g = Sr ∗ r ; %Area o f t h e e n t i r e t r a p e z o i d Area_r = area +r ∗ g ; function [ hrewind , trap_rewind , my_hdr ] = calcParams ( BW , RES , Sr , area ) % Parameters f o r r e w i n d i n g g r a d r1=sqrt ( area / Sr ) ; g1=sqrt ( area ∗ Sr ) ; my_hdr . rewinder . r1 = r1 ; my_hdr . rewinder . g1 = g1 ; % Parameters f o r t h e f i r s t r2=sqrt ( 0 . 5 ∗ area / Sr ) ; g2=sqrt ( 0 . 5 ∗ area ∗ Sr ) ; h a l f rewinding my_hdr . halfrewinder . r2 = r2 ; my_hdr . halfrewinder . g2 = g2 ; % Parameters f o r t h e T r a p e z o i d a =1.0; b = (2 .0∗ r1 −1/BW ) ; c=(area / Sr ) ; %i f i m a g i n e r y answer i f ( bˆ2−4∗a ∗ c ) < 0 error ( ' C a l c u l a t i o n E r r o r : I m a g i n a r y time . I n c r e a s e s p e c t r a l BW o r S p a t i a l ←Resolution ' ) ; else r_sol1 =(−1/2∗(b−(bˆ2−4∗a ∗ c ) ˆ ( 1 / 2 ) ) / a ) ; r_sol2 =(−1/2∗( b+(bˆ2−4∗a ∗ c ) ˆ ( 1 / 2 ) ) / a ) ; 171 PhD Thesis Sergei Obruchkov % Take t h e s h o r t e s t s o l u t i o n i f r_sol1 < r_sol2 g=Sr ∗ r_sol1 ; r=r_sol1 ; else g=Sr ∗ r_sol2 ; r=r_sol2 ; end Tf= area / g − r ; Tr= 2∗ r+2∗r1 ; end my_hdr . Area_k = Tf ∗ g ; my_hdr . trapezoid . r = r ; my_hdr . trapezoid . g = g ; my_hdr . trapezoid . Tf = Tf ; my_hdr . SpectBW = 1 / ( Tf+Tr ) ; my_hdr . Quality = Tf / Tr ; my_hdr . EffSNR = sqrt ( Tf / ( Tf+Tr ) ) ; my_hdr . S p a ti a l R e s o l u t i o n= [ num2str ( RES ) , ' cm ' ] ; % S c a l e g r a d s t o 32766 my_hdr . ampscaling = my_hdr . rewinder . g1 ; % f o r s t u p i d g = g ∗32766/ abs ( g1 ) ; g2 = g2 ∗32766/ abs ( g1 ) ; g1 = g1 ∗32766/ abs ( g1 ) ; %Dangerous b u t oh w e l l r = r ∗1 E6 ; r1 = r1 ∗1 E6 ; r2 = r2 ∗1 E6 ; Tf =Tf ∗1 E6 ; %C r e a t e waveform w i t h p r o p e r s c a l i n g hrewind = [ 0 , 0 ; r2 ,− g2 ; 2∗ r2 , 0 ] ; trap_rewind = [ 0 , 0 ; r , g ; r+Tf , g ; 2∗ r+Tf , 0 ; . . . 2∗ r+r1+Tf ,− g1 ; 2∗ r+2∗r1+Tf , 0 ] ; function grad = convtoEvenINT ( y ) %% Grad must c o n s i s t o n l y o u t o f even i n t e g e r s [ my , ny ]= s i z e ( y ) ; amp( ny ) = 0 ; dif ( ny ) = 0 ; f o r i = 1 : 1 : ny i f isnan ( y ( i ) ) amp( i ) = 0 ; dif ( i )= 0 ; else i f mod ( y ( i ) , 2 ) == 0 amp( i ) = y ( i ) ; dif ( i ) = 0 ; else i f mod ( c e i l ( y ( i ) ) , 2 ) == 0 amp( i ) = c e i l ( y ( i ) ) ; dif ( i ) = y ( i ) − c e i l ( y ( i ) ) ; else amp( i ) = f l o o r ( y ( i ) ) ; dif ( i ) = y ( i ) − f l o o r ( y ( i ) ) ; end i f mod ( y ( i ) , 2 ) == 1 amp( i ) = y ( i ) + 1 ; dif ( i ) = 1 ; 172 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics end end end end grad = int16 (amp) ; function wave_GradUP = wavetoGradUP ( wave_d ) %% Grad Updated time i s 4 us e n s u r i n g t h e wave i s p r o p e r l y s p a c e d % I n t e r p o l a t i n g v a l u e s t o t h e 1 us g r i d making s u r e time w i l l be d i v i s i b l e % i n t o 4 us i n t e r v a l [ m , n ] = s i z e ( wave_d ) ; time = wave_d ( m , 1 ) ; i f mod ( c e i l ( time ) , 4 ) ˜= 0 time = c e i l ( time ) + 4 − mod ( c e i l ( time ) , 4 ) ; else time=c e i l ( time ) ; end wave_GradUP = interp1 ( wave_d ( : , 1 ) , wave_d ( : , 2 ) , 0 : 4 : time ) ; x = 0 : 4 : time ; %f i g u r e , p l o t ( wave d ( : , 1 ) , wave d ( : , 2 ) , ' − or ' , x , wave GradUP , ' . ' ) ; 173 PhD Thesis Sergei Obruchkov 174 Dept. of Medical Physics PhD Thesis Sergei Obruchkov Dept. of Medical Physics Appendix B function dataout = reconepsi ( data , BW , BWs , BWdes , maxm , freqres , numlobes ) % I n p u t p a r a m e t e r s some i n f o you can g e t from t h e h e a d e r % This f u n c t i o n w i l l s t r i c t l y u s e manual i n p u t % L a t e r can implement p r o p e r mechanism f o r automated reco n % % Example : % % r e c o n e p s i ( data1 , 3 1 2 5 0 , 4 9 8 . 9 5 6 , 3 0 9 3 5 , 6 2 , 5 5 , 2 5 6 ) ; % % I n p u t geometry f o r r e s h a p i n g d a t a from d a t a ( e p s i r e s , avgs , p h r e s ) % i n t o data ( sres , f r e q r e s , phres ) % Need t o know t h e e p s i waveform i n f o % BW = 31250 max Sampling Bandwidth % BWs = S p e c t r a l Bandwidth % Can be measured from p h a s e b u t i t s h o u l d be % what your e p s i g r a d i e n t s were d e s i g n e d f o r % % BWdes = D e s i r e d Bandwidth f o r i n t e r p o l a t i o n % i d e a l l y BWs∗maxm % % maxm = ( i n t ) How many p o i n t s can you f i t i n t o t h e BWs % i . e . round (BW/BWs) or f l o o r (BW,BWs) % f r e q r e s = ( i n t ) number o f p o i n t s t h a t f i t on t o t h e f l a t p a r t % of the trapezoid % numlobes = maxn = number o f l o b e s , maxn∗BW/BWs < e p s i r e s % % Output s t r u c t u r e m by n % m=maxm . . . . ... . % . % . % . % m=3 . . . . ... . % m=2 . . . . ... . % m=1. . . . ... . % n=1 n=2 n=3 n = maxn % % % % T e s t i n g d a t a q u a l i t y echo each p l o t and l o o k a t peak a l l i g n m e n t % p l o t ( real ( squeeze ( dataout (5 ,: ,1:5 ) ) ) ) % f f t d a t a o u t = f f t s h i f t ( f f t n ( dataout ( : , 1 : 4 8 , : ) ) ) ; % c l e a r i pi i f nargin < 7 error ( ' See ” h e l p r e c o n e p s i ” f o r arguments ' ) ; end % I f d a t a i s l a r g e r t h e n 2D ( e p s i r e s x p h r e s ) need t o a v e r a g e [ epsires , avgs , phres ] = s i z e ( data ) ; i f ( avgs ˜= 1 ) && ( phres ˜= 1 ) data = avgepsi ( data ) ; end [ epsires , phres ] = s i z e ( data ) ; maxn=numlobes ; dt=1/BW ; dtdes = 1/ BWdes ; % V e c t o r s needed f o r i n t e r p o l a t i o n x = 0 : dt : ( epsires −1)∗ dt ; xi = 0 : dtdes : ( epsires −1)∗ dt ; 175 PhD Thesis Sergei Obruchkov Dept. of Medical Physics p ha se co r in td at a=zeros ( maxm , maxn , phres ) ; intdata = zeros ( maxm , maxn , phres ) ; f o r phase = 1 : phres data_vect = data ( : , phase ) ; intdata_vect = interp1 ( x , data_vect , xi , ' s p l i n e ' ) ; % Only need maxn∗maxm p o i n t s t o r e s h a p e % d a t a t h e r e s t can be t r a s h e d intdata ( : , : , phase )=reshape ( intdata_vect ( 1 : maxn ∗ maxm ) , maxm , maxn ) ; % Phase c o r r e c t i o n u s i n g s i n c i n p e r o l a t i o n s h i f t % Note t h a t d t d e s i s used and n o t d t s i n c e u s i n g % I n t e r p o l a t e d data f o r m = 1 : maxm f o r n = 1 : maxn p ha se co r in td at a ( m , n , phase ) = intdata ( m , n , phase ) . . . ∗exp ( i ∗2∗ pi ∗ m ∗ n ∗ dtdes ∗ BWs / numlobes ) ; end end end dataout = shiftdim ( phasecorintdata , 2 ) ; function avgdata = avgepsi ( data ) [ sres , freqres , phres ] = s i z e ( data ) ; even_data = squeeze (mean( data ( : , 2 : 2 : freqres , : ) , 2 ) ) ; odd_data = squeeze (mean( data ( : , 1 : 2 : freqres , : ) , 2 ) ) ; avgdata = even_data − odd_data ; function [ data , resi ] = ppc ( DataA , DataB ) ; % P r i n c i p a l Phase C o r r e c t i o n . Performs p h a s e r o t a t i o n o f % quadrature data without knowledge o f the phase angle . % % I /O: [ data , r e s i ] = ppc ( DataA , DataB ) ; % % DataA and DataB a r e t h e two q u a d r a t u r e c h a n n e l s g i v e n % as column v e c t o r s . % % % Henrik T o f t Pedersen , 14−Jan −2002 % i f s i z e ( DataA , 1 ) < s i z e ( DataA , 2 ) DataA = DataA ' ; end i f s i z e ( DataB , 1 ) < s i z e ( DataB , 2 ) DataB = DataB ' ; end [U,S,V] Int data resi = = = = svd ( [ DataA , DataB ] , 0 ) ; U∗S ; Int ( : , 1 ) ∗ sign (sum( Int ( : , 1 ) ) ) ; Int ( : , 2 ) ∗ sign (sum( Int ( : , 1 ) ) ) ; 176 PhD Thesis Sergei Obruchkov Dept. of Medical Physics Appendix C # 31 P and 1 H Double Tuned Coil # Solving f o r the Total Impedance of the bellow circuit # From Rectance Find Resonances and Attempt to get # parametric equations f o r L1 C1 L2 C2 in order to tune into the desired frequecies # # W1 = 2∗ Pi ∗ 1 2 7 . 8 MHz # W2 = 2∗ Pi ∗ 5 1 . 7 3 MHz restart ; # Constants # Double tuned coil Simulation See above circuit f o r Parameters and placement of ←the components # R2 is ignored as it does not contribute . . . L1 :=692 E −9; C1 : = 1 0 . 4 E −12; R1 := 0 . 5 ; L2 :=173 E −9; C2 : = 1 1 . 8 E −12; # Impedance # Z1 = L1C1R1 series impedance # Z2 = L2 | | C2 impedance Z1 := R1 + I ∗ w ∗ L1 ∗(1 −1/( w ˆ2∗ L1 ∗ C1 ) ) ; Z2 := I ∗ ( w ∗ L2 /(1−w ˆ2∗ C2 ∗ L2 ) ) ; X1 := w ∗ L1 ∗(1 −1/( w ˆ2∗ L1 ∗ C1 ) ) ; X2 :=( w ∗ L2 /(1−w ˆ2∗ C2 ∗ L2 ) ) ; Z := R1 + I ∗ ( X1 + X2 ) ; # Solving f o r Resonances where Reactance = 0 # In our case Z1+Z2 − R1 is the Reactance ; solve ( X1+X2 =0) ; eqR1 := subs ( w=2∗Pi ∗v , X1 ) : eqR2 := subs ( w=2∗Pi ∗v , X2 ) : eqRsum := subs ( w=2∗Pi ∗v , X1+X2 ) : plot ( [ eqR1 , eqR2 , eqRsum ] , v=1E7 . . 2 E8 , y=−5E2 . . 5 E2 ) ; # Solving the Equation f o r the Previous Solution # The resonances are in Hz and are dead on as expected from the RF Simulation ( see ←Schematic Above ) evalf ( solve ( C2 = ( w ˆ2∗ L1 ∗ C1−1+w ˆ2∗ L2 ∗ C1 ) / ( w ˆ2∗ L2 ∗ ( w ˆ2∗ L1 ∗ C1 −1) ) , w ) / ( 2 ∗ Pi ) ) ; # Explicit Solution and Numerical Calculations solve ( C2 = ( w ˆ2∗ L1 ∗ C1−1+w ˆ2∗ L2 ∗ C1 ) / ( w ˆ2∗ L2 ∗ ( w ˆ2∗ L1 ∗ C1 −1) ) , w ) ; # Explicit Solutions f o r the Resonant Frequencies # W1 = 2∗ Pi ∗ 1 2 7 . 8 E6 MHz # W2 = 2∗ Pi ∗ 5 1 . 7 3 E6 MHz W1 :=1/2∗ sqrt ( 2 ) ∗ sqrt ( C2 ∗ L2 ∗ L1 ∗ C1 ∗ ( L2 ∗ C1+C2 ∗ L2+L1 ∗ C1+sqrt ( L2 ˆ2∗ C1 ˆ2+2∗ L2 ˆ2∗ C1 ∗ C2+2∗←L2 ∗ C1 ˆ2∗ L1+C2 ˆ2∗ L2 ˆ2−2∗ C2 ∗ L2 ∗ L1 ∗ C1+L1 ˆ2∗ C1 ˆ 2 ) ) ) / ( C2 ∗ L2 ∗ L1 ∗ C1 ) ; W2 :=1/2∗ sqrt ( 2 ) ∗ sqrt ( C2 ∗ L2 ∗ L1 ∗ C1 ∗ ( L2 ∗ C1+C2 ∗ L2+L1 ∗ C1−sqrt ( L2 ˆ2∗ C1 ˆ2+2∗ L2 ˆ2∗ C1 ∗ C2+2∗←L2 ∗ C1 ˆ2∗ L1+C2 ˆ2∗ L2 ˆ2−2∗ C2 ∗ L2 ∗ L1 ∗ C1+L1 ˆ2∗ C1 ˆ 2 ) ) ) / ( C2 ∗ L2 ∗ L1 ∗ C1 ) ; # L1 Variable is fixed since it is determined by the coil geometry # L2 Is fixed f o r calculation purposes : L1 := evalf ( 1 / ( ( 2 ∗ Pi ∗129 E6 ) ˆ 2 ∗ 2 . 2 E −12) ) ; L2 := L1 / 2 ; M := solve ( { W1=2∗Pi ∗ 1 2 7 . 8 E6 , W2=2∗Pi ∗ 5 1 . 7 3 E6 } , [ C1 , C2 ] ) ; M[1]; # Steping through L2 in a loop # NOTE : after 320 nH Solutions are imaginarry might need the other part of the ←equations ? f o r n from 1 to 13 do L2 := 100 E−9+20E −9∗(n −1) : solve ( { W1=2∗Pi ∗ 1 2 7 . 8 E6 , W2=2∗Pi ∗ 5 1 . 7 3 E6 } , [ C1 , C2 ] ) : end do ; 177 PhD Thesis Sergei Obruchkov Dept. of Medical Physics # # Multiple Breaks # Now we have n breaks with X2 =Sum ( Xn ) X2 :=( w ∗ L2 /(1−w ˆ2∗ C2 ∗ L2 ) ) ; L2 :=160 E −9; C2 : = 1 6 . 9 E −12; n :=2; Ln := L2 / n ; solve ( X2 = n ∗ ( w ∗ Ln /(1−w ˆ2∗ Cn ∗ Ln ) ) ) ; # The rule seems to be L / n C ∗ n where n is the number of breaks 178