Using GATE to understand performance of a full-torso PET scanner

Transcription

1 University of Pennsylvania Using GATE to understand performance of a full-torso PET scanner Varsha Viswanath 1 Margaret E. Daube-Witherspoon 1, Matthew E. Werner 1, Suleman Surti 1, Andreia Trindade 2, Pedro Rodrigues 2, Austin R. Pantel 1, Amy E. Perkins 2, and Joel S. Karp 1 1 University of Pennsylvania 2 Philips Healthcare IEEE MIC 2017

2 The full-torso scanner will allow for low-dose adult imaging, fast pediatric imaging, full-torso dynamic imaging with novel tracers, and imaging of low β+ fraction isotopes

3 The full-torso scanner will allow for low-dose adult imaging, fast pediatric imaging, full-torso dynamic imaging with novel tracers, and imaging of low β+ fraction isotopes Full-torso scanner 23-cm scanner Data courtesy of Drs. Austin Pantel and Robert Mach, University of Pennsylvania

4 The full-torso scanner will allow for low-dose adult imaging, fast pediatric imaging, full-torso dynamic imaging with novel tracers, and imaging of low β+ fraction isotopes 18 F has a positron fraction of %

5 Simulating a full-torso scanner

6 16-cm Clinical Philips Vereos 23-cm Prototype One-Ring Scanner 70-cm Full-Torso Three-Ring Scanner We modeled the 16-cm Philips Vereos scanner in GATE and extended it to 23 cm and to a 70-cm full-torso scanner 320 ps Timing Resolution 10 % energy resolution 4 x 4 x 20 mm 3 crystals

7 NEMA Experiments Sensitivity Count Rate Spatial Resolution Image Quality 4:1

8 Our Computing Resources Node 2 24 GB RAM 12 CPU cores Node 3 24 GB RAM 12 CPU cores Node 4 24 GB RAM 12 CPU cores Totals 76 CPU cores Node 1 2 GATE runs per Node 5 12 GB RAM 12 CPU cores Node 6 12 GB RAM 12 CPU cores Node 7 32 GB RAM 16 CPU cores core Up to 152 GATE runs per batch Each node includes dual GHz Nehalem-class Xeon CPUs

9 Running the 3-minute NEMA IQ Simulation In General, GATE runtimes were linear up to root file sizes of 1 GB A 3-minute GATE scan required ms simulations

10 Running the 3-minute NEMA IQ Simulation 100 ms scan took 44 min: 23-cm scanner 75 min: 70-cm scanner Total Runtimes 8.5 hours: 23-cm scanner 15 hours: 70-cm scanner

11 Simulating a voxelized XCAT 2.0 anthropomorphic phantom

12 Simulation Plan [ 18 F] FDG Patient Scan Adult Female 5 5 (165 cm) 165 lbs (75 kg) List of Important Organs Body (i.e. Fat) Other Bone Lung Breast (F) Muscle Pancreas Gut Spleen Bladder Blood Pool Uterus (F) Spine Bone Liver Kidney Brain (WM) Myocardium Brain (GM) Total = 17 organs

13 Filling the XCAT anthropomorphic phantom with an [ 18 F]FDG distribution #1: Atlases in literature

14 Filling the XCAT anthropomorphic phantom with an [18F]FDG distribution [ 18 F] FDG Patient Scan Dr. Austin R. Pantel #2: Physician input using patient data N = 10

15 Filling the XCAT anthropomorphic phantom with an [18F]FDG distribution ActivityRange.dat #3: Voxelize activity Each voxel is discrete size RAM size coarse dist n Body, Other Bone Lung Breast(F) Muscle Pancreas Gut Spleen, Blood Pool, Uterus (F) Spine Bone Liver, Kidney, Brain (WM) Myocardium Brain (GM) Bladder

16 Filling the XCAT anthropomorphic phantom with an [18F]FDG distribution #4: Crop the Image down to minimize RAM usage - Save as analyze - Set datatype to 16 (floating point) 192 x 192 x x 100 x 230

17 Converting the XCAT Attenuation Image for GATE µ map w/22 organs Index map w/17 organs Segment Crop

18 GATE simulations of the XCAT phantom on a fulltorso scanner take a LOT of time GATE V8.0 3 weeks later

19 To include small lesions in our phantom, we simulated them separately and embedded them Lesion w/ 3:1 contrast fill with 2x the local background activity GATE V min later

20 Total List Mode data Total List Mode data Merged list files and subsampled data to study the dose lowering effects of the full-torso scanner whole 1/2 1/4 1/8 1/16 Lesion List Mode data x10 x10 x10 FDG woman List Mode data x10 LM TOF OSEM 4 x 25 subsets This was done as: 1 bed pos 70-cm scanner 5 bed pos 23-cm scanner x10 Mean (CRC) Std (CRC) x10 x5 x5

21 The full-torso scanner maintains measurement accuracy and precision at lower imaging doses Patient dose can be lowered by 2x-4x CRC %SD

22 Summary 1. How I run simulations on our cluster - Parallelizing GATE into many short runs 2. How I process the XCAT phantom for GATE simulations - Phantom was filled using data from literature and on-campus imaging studies - Attenuation and Activity images were segmented into discrete organs - Simulations were run using analyze images 3. Future Directions - Planning simulations of dynamic datasets to understand how the improved sensitivity of a full-torso scanner will impact dose and temporal sampling for improved kinetic parameter estimation

23 Amy E. Perkins Pedro Rodrigues Andreia Trindade Funding: NIH R01-CA NIH R01-CA Philips Research Agreement Joel S. Karp Margaret E. Daube- Witherspoon Suleman Surti Matthew E. Werner Austin R. Pantel