Room: AAPM ePoster Library
Purpose: High quality, low-dose megavoltage cone-beam CT (MV-CBCT) may be important for adaptive radiotherapy applications like online re-planning. Digital prototyping of novel imager designs for optimization using Monte Carlo simulation (MCS) avoids hardware implementation cost but requires prohibitively long computation times. In this study, a novel strategy is presented to accelerate the MCS of MV-CBCT using a graphic processing unit (GPU).
Methods: A novel MV-CBCT simulation framework that generates a series of phantom projections from a single simulation run is developed. Primary photons are generated from a beam source and saved in a particle batch. Since the beam source is independent from Linac gantry rotation, the same batch is rotated round the phantom volume to simulate projections at different angles. To eliminate fixed patterns among the projections and ring artifacts on reconstructed images, the particle batch is randomly rotated around its central axis. A GPU-based Geant4 code is incorporated into the framework to simulate photon transport through the phantom volume. The FastEPID method, which accelerates the simulation of MV image by generating projection from pre-calculated imager response, is modified and integrated into the framework. Phantom projections at the same angle but generated from different batches are accumulated to form the final image. A standard FDK algorithm is used for reconstruction with the projections extracted from the simulation.
Results: The proposed GPU-based MV-CBCT simulation framework was validated against measurement and compared to CPU simulation for a Catphan 604 phantom at beam energies of 2.5 MV, 6 MV, and 6 MV FFF. The reconstructed image quality was comparable between GPU simulation, CPU simulation and measurement. The GPU simulation was accelerated by factors of 900–1900 times, compared to the CPU-based simulation.
Conclusion: The proposed GPU-based framework is suitable for accurate MV-CBCT simulation and will accelerate MV imager design optimization and other related applications.
Funding Support, Disclosures, and Conflict of Interest: NIH/NCI R01CA188446