(Tuesday, 7/31/2018) 11:00 AM - 12:15 PM
Room: Room 202
Purpose: Monte Carlo (MC) method is widely used to study PET system for hardware/prototype designs, image reconstruction, artifact reduction and applications in hadron-therapy. Existing packages suffer from low computational efficiency. This study develops a GPU-based highly efficient and accurate MC tool, gPET.
Methods: gPET was developed on CUDA computation platform. The simulation process includes positron decay, photon transport inside a voxelized phantom, and signal detection in a crystal detector modeled by parameterized geometry. The detector was formed in three levels: panel, module, and crystal scintillator. A user can selectively tally multiple quantities such as intermediate photon phase space file (PSF), or energy, interaction-type, local-coordinates, global-time and crystal-index for events inside the detector. We performed three comprehensive tests to evaluate the performance of gPET, including: 1) comparing PSF calculated using gPET and Geant4, 2) comparing PET signals calculated using gPET and GATE and 3) comparing reconstructed images using signals calculated by gPET and GATE.
Results: 1) 5e6 history was simulated and relative yield difference (RYD) at 511 keV peak energy was 0.34%. 2) 6.5e6 events were generated in the detector after 4e6 source history in gPET. Compared to GATE, RYD for module (6.5x6.5 cm²) with events larger than 1e3 was 2.7%, while that for crystal (0.2x0.2 cm²) with events larger than 5e2 was 3.3%. 3) 2e6 history was simulated. RYD for the reconstructed images with intensity larger than 1e-3 was 0.2%. Computation time of gPET was 1.8 microsecond/history in cases 1) and 2) on a GeForce GTX TITAN Black card, in comparison with 35 millisecond/history for the same cases in Geant4 and GATE.
Conclusion: gPET is an accurate and efficient tool for complex PET simulations. The package is open to research community.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
