Fifty-seventh annual meeting of the American association of physicists in medicine
TH-EF-BRD-02: An Analytic Linear Accelerator Source Model and Automatic Source Commissioning for GPU-Based Fast Monte Carlo Dose Calculation
GPU-based Monte Carlo (MC) dose calculations have become available. Linac source modeling for it is still a problem to be addressed. The model should be carefully designed for sufficient accuracy, ease of automatic commissioning, and more importantly suitability for parallelization. This abstract presents an analytical source model for GPU-based MC dose engines with these factors considered.
Analytical source models are preferred for GPU comparing to phase-space-file based models due to the avoidance of loading and transferring large data. In our model, we introduced a phase-space-ring (PSR) concept. A PSR contains particles of the same type and close in energy and radial distance on a plane above jaws. We parameterized probability densities of particle location, energy, and direction for three sub-source types: primary photon PSR, scattered photon PSR and electron PSR. Model parameters were derived by analyzing reference phase-space files. To incorporate this model into our GPU dose engine, we designed an efficient sampling scheme to reduce GPU thread divergence. Automatic commissioning was achieved by solving an optimization problem to adjust PSR weight.
Our model built for Varian Truebeam 6MV photon beam was able to represent the reference phase-space. Comparing open-field doses in water computed using our model and the phase-space file, average distance-to-agreement(DTA)≤1mm for high-gradient regions, root-mean-square(RMS) difference≤1.1% for low-gradient regions and output factor difference≤0.5% were observed. Good agreements were seen in IMRT cases. Our model improved efficiency by 1.70–4.41 times compared to phase-space-based source model. It took ∼20sec to perform commissioning. Compared to uncommissioned model, DTA was reduced from 0.04–0.28cm to 0.04–0.12cm for depth dose at build-up region, and RMS from 0.32%–0.67% to 0.21%–0.48%, 0.31%–2.0% to 0.06%–0.78% and 0.20%–1.25% to 0.10%–0.51% for depth dose after build-up, lateral profiles of inner beam and outer beam, respectively.
The analytic source model achieved high accuracy, efficiency, and suitability for GPU-based dose calculations.