Compton Scattering as a Tool for Medical Imaging Harris Kagan Ohio State University UCLA April 14, 2005 Outline of Talk • Introduction • Compton Imaging • PET • Prostate Probe • Summary UCLA April 14, 2005 The CIMA Collaboration Compton Imaging for Medical Applications - Collaborators • The Ohio State University Columbus, Ohio USA • The University of Michigan Ann Arbor, Michigan USA • CERN Geneva, Switzerland • Instituto de Física Corpuscular Valencia, Spain • Institut Jozef Stefan Ljubljana, Slovenia UCLA April 14, 2005 Introduction Early Detection of disease is essential in: – prostate cancer • Most common diagnosed non-skin cancer in US • Second leading cause of cancer-related deaths in Western male population • One in 6 men will develop prostate cancer; 1 in 26 will die • American man is 33% more likely to develop prostate cancer than an American woman to get breast cancer – breast cancer • • • • Most common cancer among women One in 9 women will develop breast cancer; 1 in 27 will die Number of new cases is growing at ~2% rate Smallest detectable cancer 2-3 mm Early detection photon imaging UCLA April 14, 2005 Introduction Many Imaging Tasks: Detection Classification Choose between two: Choose among many Estimation Choose a value from a continuum: •Tumor present •Radiotracer uptake in a region •Tumor absent •Exact location of tumor UCLA April 14, 2005 Introduction Three main interaction of photons with matter •Photoelectric effect •Compton Effect •Pair Production UCLA April 14, 2005 Source Energy [keV] Half-life 99mTc 140.5 6 hr 111In 171, 245 2.8 days 131I 364 8 days 22Na, 18F, 11C, 15O 511 950 days, 1.8 hr Introduction Early Detection with Imaging SPECT - Single Photon Emission Computed Tomography UCLA April 14, 2005 Introduction For mechanical collimators the tradeoff between detection efficiency and spatial resolution is unfavorable. object S : collimator efficiency Rc : collimator resolution collimator S Rc2 detector UCLA April 14, 2005 ~10-15 mm FWHM spatial resolution ~detection efficiency of 10-4 for body imaging (99mTc) Introduction About Absorbing Collimation • In the best case efficiency is only a few photons detected for 10,000 emitted • Performance of absorbing collimation decreases rapidly with increasing energy - Photoelectric absorption coefficient changes as ~ Z4 / E3 • Typical collimator for 140 keV (Tc99m)– 9mm FHWM @ 10cm, 2.3e-4 • Typical for 364 keV (I-131) – 12mm FWHM @ 10cm, 1e-4 • Radionuclides with even a small fraction of higher energy radiation will result in severely compromised imaging performance UCLA April 14, 2005 Introduction Existing Approaches for Improvement • Larger detector area (multiple head, ring geometry) sensitivity but price & limited improvement Non-uniform sampling (fan beam, cone beam collimators) sensitivity & resolution but FOV & distorted images Multiplexing (multiple pinhole arrays, coded aperture) high sensitivity but computationally costly & extra noise propagation from one part of image to another part UCLA April 14, 2005 Compton Imaging • Electronic collimation – uncouple resolution and efficiency • Two detectors in coincidence - Near field operation - Photon scatters in first detector – tag, efficiency - Photon absorbed in second – no collimation, efficiency • Reconstruction of scattering angle based on kinematics -Measure interaction points and energy in both detectors • Good performance from 140-511 keV • Should enable excellent images where conventional techniques only partially successful UCLA April 14, 2005 Compton Imaging Electronic Collimation 1st Det Compton Camera 2nd Det 1 0 ee- 1 - 1 1 (1-cos) E1 E0 = 511keV E0: incident photon energy E1: scattered photon energy : scattering angle 1st det: solid state detector 2nd det: scintillation detector UCLA April 14, 2005 Source Image Plane 1st Detector 2nd Detector Scattered -Rays Compton Imaging - Goals • Investigate the performance of PET –Goal Detection efficiency – Spatial resolution • Investigate the performance of Compton probe – Influence of non-prostate background activity – Prostate tumor detection UCLA April 14, 2005 Compton Imaging - History 1922 1968 1974 1981 1983 1987 1993 1996 Compton scattering (Compton) Compton telescope (White) nuclear medicine application (Todd et. al.) Compton image reconstruction (Singh et. al.) first prototype Compton camera (Singh) multiple Compton method (Kamae et. al.) ring Compton camera (Martin et. al.) Compton-SPRINT (Rogers et. al.) Also Astrophysics: COMPTEL, TIGRE, MEGA, … UCLA April 14, 2005 Applications of Small High Resolution PET Metabolic studies with small animals Studies with genetically modified mice Radiotracer development Sub-millimeter resolution is required for small organs (thyroid 1~6mg, adrenal 3~40mg) and functional regions of brain, kidney, etc. UCLA April 14, 2005 General Principles of PET Patient injected with drug having + isotope •Scintillation Detector Ring •(continuous or segmented) Drug localized in patient Radioisotope decays and emits + + annihilates with e- from tissue and forms back-to-back 511 keV photon pairs •Patient •e- 511 keV photon pair detected with timing coincidence window Annihilation position lies on line of response (LOR) defined by detector pair •Scintillation site UCLA April 14, 2005 •+ Role of Image Reconstruction Scanner with some set of Line Of Responses Projection data collected along a “tube” or line of response (LOR) Scanner LOR data Image Reconstruction Algorithm Patient Display Effective field of view (FOV) Assumption (events in LOR) constant (activity concentration in patient) LOR UCLA April 14, 2005 Compton PET Concept •BGO •detector •Si •detector •Three Major Coincidence Events •Si-Si : • Very High Resolution • •Si-BGO : High Resolution •Si-Si •Si-BGO •BGO-BGO UCLA April 14, 2005 •BGO-BGO : • Conventional PET Resolution Sensitivity and Spatial Resolution • Si-SI Si-BGO •Si-Si • Sensitivity: 1.0 % • *FWHM = 230 m •Si-BGO • Sensitivity : 9.0% • *FWHM = 790 m •BGO-BGO 1D Profiles •BGO-BGO • Sensitivity: 21.0 % • *FWHM = 1.45 mm •* Not including •effects of annihilation photon acolinearity and positron range UCLA April 14, 2005 •Image reconstruction: FBP Overall Spatial Resolution Si-BGO Si-Si Event Geometric BGO-BGO Spatial Resolution (mm FWHM) Geometric Overall + Acolinearity F-18 C-11 N-13 O-15 Si-Si 0.234 0.241 0.393 0.443 0.492 0.553 Si-BGO 0.788 0.816 1.062 1.261 1.419 1.742 BGO-BGO 1.452 1.458 1.977 2.270 2.490 3.069 UCLA April 14, 2005 Spatial Resolution UCLA April 14, 2005 Compton Scatters in Scintillator Conventional PET • Scintillator •~ 1 cm Multiple interactions Energy deposited over a volume ~ 1cm mean path Penetration into crystal ring blurs measured position due to depth of interaction (DOI) uncertainty Resolution ~ 1 – 2 mm UCLA April 14, 2005 Scatter and DOI Uncertainty in BGO PET •- Using Monte Carlo simulations •- BGO PET ( 17.6 cm I.D. 16 cm length segmented with 3 mm x 3mm x 20 mm crystals) •- Point sources at 0, 3, 6, 9, 12, 15, and 18 mm from center of FOV •- Filtered back projection (FBP) method •True first interaction position UCLA April 14, 2005 •Centroid of scattered E distribution •DOI uncertainty included in both images Annihilation Photon Acolinearity Deflection for Acolinearity Distribution of Deflection Angles 511 keV photon 180 e from 10 million photon pairs Angular Uncertainty : Gaussian distribution Mean = 0 degree FWHM = 0.5 degree UCLA April 14, 2005 [by Evans and Harrison] Compton Kinematics Compton Kinematics Reduction of scatter, random, BGO ESi Si and misclassified events Si BGO Angular Uncertainty Factors Doppler Broadening Detector Element Size Si pad: 0.3 mm x .3 mm x 1 mm BGO crystal: UCLA April 14, 2005 3 mm x 3 mm x 20 mm EBGO Energy Resolution Photon Acolinearity Blurring due to Positron Range Intrinsic F-18 C-11 N-13 O-15 Si-Si Si-BGO BGO-BGO UCLA April 14, 2005 All images include photon acolinearity. Each image size is 10 mm x 10 mm First Detector UCLA April 14, 2005 2D Images of Simulated Multiple Cylinder Sources Sources Si-Si (160k) - F-18 Cylinder Sources : 1,2,4,6,8, and 10 mm dia, 10 mm long -Centers of Each Source : 1cm from center of FOV 4 cm x 4 cm -Filtered back projection (FBP) -Images reconstructed with different resolution and sensitivity data UCLA Si-BGO April 14, 2005 (1.4M) BGO-BGO (3.1M) Multi-resolution Image Reconstruction - Imaging system generates data having wide different spatial resolution and sensitivity. - FBP does not account for the different information in the three datasets. - How to combine these image ? Si-Si (160k) FWHM: 0.34 mm UCLA April 14, 2005 Si-BGO (1.4M) 1.07 mm BGO-BGO(3.1M) 1.95 mm All from FBP Image Combined with ML-EM - Maximum likelihood Expectation Maximization (ML-EM) - Iteration number = 200 4 cm x 4 cm Si-Si (160k) + Si-BGO (1.4M) + BGO-BGO (3.1M) UCLA April 14, 2005 A Prototype Experimental Setup BGO detector (non-position sensitive) Lead shielding and slice collimation Source turntable Silicon detector in plastic box Readout Board for BGO Intermediate Board for Si readout UCLA April 14, 2005 Silicon and BGO Detector Silicon detector BGO detector VATAGP3 HAMAMATSU PMT R2497 4.5 cm 2.2 cm and 1 mm thick 3216 (512) pads, 1.4 mm 1.4 mm pixel size VATAGP3 : 128 slow shapers (VA part, preamp) 128 fast shapers (TA part, trigger) UCLA April 14, 2005 5.3 cm 5 cm and 3 cm thick 84 array, 12.5 mm 5.25 mm crystal size Performance of Silicon Detector UCLA April 14, 2005 Baseline offset and Noise Gain and FWHM Energy Resolution of Silicon Detector Am-241 (59.5 keV) FWHM = 1.49 keV (2.5 %) UCLA April 14, 2005 Tc-99m (140.5 keV) FWHM = 1.39 keV (0.99%) Pb K1 = 74.969 keV, K2 = 72.804 keV, K1 = 84.936 keV, and Compton edge = 49.8 keV Energy Resolution of BGO Detector Tc-99m (140.5 keV) FWHM = 65.5 keV (46.6 %) UCLA April 14, 2005 Na-22 (511 keV) FWHM = 116.6 keV (22.8%) Compton edge : 340 keV for 511 keV Compton edge : 1061.7 keV for 1274.5 keV Detector Alignment Silicon detector Lead shielding Silicon detector Tungsten Slit Source turntable Laser beam UCLA April 14, 2005 Compton PET System Block Diagram BGO BGO Si Si BGO BGO CFD CFD Coincidence Unit VME DAQ Trigger line VME DAQ Computer Listmode data (E, position) UCLA April 14, 2005 Lines of Response (Alignment) UCLA April 14, 2005 Hit Map UCLA April 14, 2005 Images of Two Point Sources Compton PET MicroPET R4 Source F-18 in glass capillary tubes 5 F-18 5 4 4 1.1~1.2 mm 3 0.4 mm 3 2 2 1 1 Glass wall 0.2 mm 0 0 cm 1 UCLA April 14, 2005 2 3 4 5 0 0 cm 1 2 3 ML-EM Image reconstruction with Si-Si coincidence events 4 5 Images of Two Point Sources 5 5 Compton PET MicroPET 4 1.1~1.2 mm 4 3 2.0 mm 3 2 2 F-18 1 0 0 cm 1 2 3 4 5 Glass wall 0.2 mm 1 0 0 cm 1 2 3 4 Source FWHM = 1.14, 1.11 mm UCLA April 14, 2005 FWHM = 1.72, 1.61 mm 5 Intrinsic PET Resolution Image Resolution 1.11~1.14mm FWHM Intrinsic Resolution ~ 700 m FWHM Intrinsic Resolution Source Size = Image Resolution UCLA April 14, 2005 Intrinsic PET Resolution F-18 0.254 mm 0.127 mm Needle 25G (ID = 0.254 mm, OD = 0.5mm, SS_steel wall = 0.127 mm) SS_steel wall 5 4 Image Resolution = 700 m FWHM 3 2 1 0 0 cm 1 UCLA April 14, 2005 2 3 4 5 Resolution Uniformity 5 5 4 4 3 3 2 2 1 1 0 0 cm 1 2 3 4 F-18 in capillary tubes (I.D. 1.1 ~ 1.2 mm) Resolutions: ~ 700 µm FWHM 5 0 0 5cm 1 2 3 4 5 1 2 3 4 5 4 3 2 1 UCLA April 14, 2005 0 0 cm Compton PET Summary In order to achieve sub-millimeter spatial, a small animal PET based on Compton camera concept was proposed. Simulation results demonstrated sub-millimeter spatial resolution of the Compton PET (0.4 mm FWHM from Si-Si and 1.0 mm FWHM from Si-BGO). Image reconstruction method using MLEM was developed to properly combine images having different resolutions and sensitivities. Experimental results with a prototype setup using 1.4 mm x 1.4 mm x 1 mm silicon pads verified very high resolution (700 m FWHM). UCLA April 14, 2005 Compton Prostate Probe The Big Question: How well can a compton camera work relative to a conventional collimated single photon system? UCLA April 14, 2005 Compton Probe Monte Carlo Studies Prob e Prostate Geometry set-up for “voxel-man” simulation UCLA April 14, 2005 • Human phantom – “Zubal” male torso phantom with 4mm3 voxel size mated with EGS4 • Internal probe and 2nd detectors • Radioisotopes – Tc-99m, In-111, I-131, and positron emitters Compton Probe Monte Carlo Studies Counting Efficiency Probe Efficiency 10% FWHM Energy Window 0.20% 0.18% 0.16% 0.14% 0.12% 0.10% 100 200 Tc99 m 300 In111 400 I131 (ProstaScint) Gamma Photon Energy UCLA April 14, 2005 500 e+ 600 Compton Probe Reconstruction List-mode Likelihood Reconstruction Images of “M” at 100 Iterations 141kev 171kev 245kev 364kev 511kev 4mm Point-to-Point, 1cm from probe UCLA April 14, 2005 Compton Probe UCLA April 14, 2005 Compton Probe UCLA April 14, 2005 Compton Probe B cts/ Prostate cts Background from Other Organs 0.25 0.2 0.15 0.1 0.05 0 0 100 200 300 400 Gamma Photon Energy Target Activity Bkgd. Activity UCLA April 14, 2005 liver (0.009) kidney (0.06) spleen (0.1) bladder (1.0) 500 600 Compton Probe Planar Images of Prostate with Tumors Prostate with 4mm tumors 171keV (245k counts) Tumor 2 15:1 conc. Tumor 3 30:1 conc. Tumor 1 10:1 conc. 80 iterations UCLA April 14, 2005 100 iterations #1: 0.8cm away, 10:1 T/P concentration ratio #2: 1.2cm away, 15:1 T/P concentration ratio #3: 1.6cm away, 30:1 T/P concentration ratio Compton Probe Relative Uptake of In-111 Prostascint Organ Relative Uptakes Prostate 1.0 Liver 2.0 Blood 1.5 Bone 0.7 Kidney 1.0 Spleen 1.0 Bladder 0.6 Rectum 0.4 Testes 0.6 Averaged from three In-111 Prostascint SPECT scans UCLA April 14, 2005 Compton Prostate Probe Prostate Probe Reconstruction 171 keV Prostate w/tumor Prostate only UCLA April 14, 2005 245 keV Compton Prostate Probe Prostate Probe Reconstructions 245 keV only, 1.2 million events, 8mm lesion 5:1 w / tumor bkgd UCLA April 14, 2005 10:1 15:1 20:1 SPECT SPECT Reconstructions 5:1 w / tumor bkgd UCLA April 14, 2005 10:1 15:1 20:1 Compton, SPECT Performance Performance 8mm Tumor Area Under ROC Curve: 8 mm Tumor 1 Area Under Curve (AUC) 0.95 Compton Probe 0.9 0.85 0.8 0.75 0.7 Conventional SPECT 0.65 0.6 0.55 0.5 1 UCLA April 14, 2005 1.5 2 2.5 3 3.5 4 Tumor-to-background Ratio 4.5 5 Compton, SPECT Performance Performance 5mm Tumor Area Under ROC Curve: 5 mm Tumor 1 Area Under Curve (AUC) 0.95 0.9 0.85 Compton Probe 0.8 0.75 0.7 0.65 Conventional SPECT 0.6 0.55 0.5 1 UCLA April 14, 2005 2 3 4 5 6 7 8 Tumor-to-background Ratio 9 10 Conclusions – Prostate Probe • Compton probe has overall 0.16% counting efficiency for Gamma energies ranging from 141kev to 511kev • Compton probe has FWHM resolution <2.9mm; best (2.1 mm) at 364 keV • Simulations of probe exhibited 3.5x increase in SNR over conventional Prostascint SPECT • Compton performance improves dramatically with energy • Tumor much easier to visually locate than with conventional SPECT • Useful over a wide range of energies (140 >511 keV) • Small device in construction UCLA April 14, 2005 Summary • Compton Imaging has been demonstrated to have applicability in many areas • High Resolution PET with < 1mm FWHM has been achieved – Presently constructing full barrel – Clinical tests? • Prostate Probe works well in simulations – Presently constructing new prototype • Compton Scintimammography is still an open question – simulations in progress • Many open questions need attention UCLA April 14, 2005 Extra Slides UCLA April 14, 2005 Compton Scintimammography Requirements: • Reliable detection of small (<1cm diameter) lesions • Ability to operate at many energies (inc. 511 keV) for tracer flexibility UCLA April 14, 2005 The Compton Imaging Probe “Near-field” operation provides the following advantages • High spatial resolution • High counting efficiency • Useful for 140 keV and higher energies • Relatively low-cost UCLA April 14, 2005 2nd Detector 1st Detector Object Object is placed close to 1st detector Uncertainty increases with distance The Contenders: SPEM • • • • • • UCLA April 14, 2005 Simple Collimators close to breast have high resolution Can trade some resolution for efficiency Can see lesions in chest wall using angled collimators Flexible configurations Efficiency very low – long imaging times The Contenders: PEM • High efficiency and resolution • Tracers use “elements of life” • Sensitivity drops off toward edges of FOV • Difficult to see lesions close to the chest wall UCLA April 14, 2005 Compton Emission Mammography High efficiency and resolution • “Sees” lesions throughout breast with high efficiency • Capable of operating over wide energy range with good performance • Coded data – performance advantage is not raw efficiency gain • Influenced by background in this configuration • Computationally intense reconstruction • UCLA April 14, 2005 CEM in more detail • Resolution and efficiency vs. incident energy • Influence from background activity • Preliminary lesion detection performance. UCLA April 14, 2005 Configuration and Experiments 10cm x 10cm silicon detectors of various thicknesses • 2 40cm x 40cm scintillation cameras • Data generated by attaching 2 10cm x 10cm x 5cm breast phantoms to “Voxelman”/EGS4 Monte Carlo code • ~1.3 M events / realization • UCLA April 14, 2005 Spatial Resolution 4 Spatial Resolution (mm FWHM) 3.5 2.5 cm 3 2.0 cm 2.5 1.5 cm 2 1.0 cm 1.5 1 100 0.5 cm 150 200 250 300 350 Energy (keV) UCLA April 14, 2005 400 450 500 550 Efficiency 0.045 10mm 0.04 0.035 Absolute Efficiency 0.03 5mm 0.025 0.02 0.015 2mm 0.01 0.005 1mm 0 100 150 200 250 300 350 Energy (keV) UCLA April 14, 2005 400 450 500 550 Efficiency -5 mm thickness 0.05 0.5 cm 0.045 0.04 Absolute Efficiency 1.0 cm 0.035 1.5 cm 2.0 cm 0.03 2.5 cm 0.025 0.02 0.015 100 150 200 250 300 350 Energy (keV) UCLA April 14, 2005 400 450 500 550 Lesion Reconstruction Experiments •5 cm •8mm / 10:1 •4-plane •8mm / 8:1 •3-plane •5mm / 15:1 •5 mm •8 mm •Extra plane for 4 plane reconstruction •Note improvement in lesion contrast with better modeling UCLA April 14, 2005 Reconstruction – no background •140 keV •171 keV •245 keV •8mm •5mm •1.2 M events UCLA April 14, 2005 •364 keV •511 keV Reconstruction with background •140 keV •171 keV •245 keV •364 keV •8mm •5mm •1.2 M events •Liver:Heart:Normal Breast 4:2:1 UCLA April 14, 2005 •511 keV Background: Principle of Compton Camera Source Image Plane • Detection coincident events between two detectors • Compton scatter equation relates scatter angle and Eo and Ere 1st Detector cos 1 2nd Detector Scattered - Rays UCLA April 14, 2005 511 511 E 0 E 0 E re • Photon direction is determined within conical ambiguity “slab-man” Monte Carlo Simulation Planar CZT came ra • Human phantom – 40x40x20cm3 slab tissue material with soft • Internal probe 20cm 30cm Silicon detector Human phantom 40cm 40cm UCLA April 14, 2005 Planar CZT came ra – 1x4x1cm3 silicon with 1mm3 detector elements • External 2nd detectors – 40x40x2cm3 Cadmium Telluride (CZT) Zinc • Radioisotopes – Point sources of Tc-99m, In111, I-131, and positron emitters – Located 1cm away from the front face of silicon detector Applications Use Silicon detectors for spatial resolution BGO detect or BGOBGO 5 Compton scattering in silicon5 Si detect Si-or Si SiBGO 4 Compton PET 3 3 2 2 1 0 5 cm 0 Silicon detector Lead shielding 4 1 2 3 4 1 MicroPET R4 0 5 cm 1 2 3 4 0 Silicon detector Tungsten Slit Source turntable UCLA April 14, 2005 Laser beam Compton Probe Probe-Gamma Camera Comparison Efficiency and resolution for In111 Imaging Distance Compton Probe High-Sensitivity Collimator High-Resolution Collimator UCLA April 14, 2005 10 cm Efficiency 1.8e-3 1.11e-4 4.00e-5 Resolution 2.47mm 15.9mm 10.5mm Monte Carlo Simulations • Used EGS4/Voxelman simulation for prostate probe data. 10 x 1.3 million event realizations generated. • Realizations were pooled and another 9 generated by resampling with replacement (“bootstrapping”) • SIMND with Voxelman and model for hi res / medium energy collimator used to generate SPECT data • 25 realizations generated having counts same as observed in Prostascint SPECT scans at UM • NPW matched filter applied using templates from tumoronly images, SNR of filter output estimated UCLA April 14, 2005 SPECT Reconstruction 171 + 245 keV (20% window), Hi Res / Med Energy collimator Without Tumor Prostate 8 mm dia Tumor Only UCLA April 14, 2005