Implementation of An Octree-Based CT Compression for GPU-Based Monte Carlo Dose Calculations
Presentations
PO-GePV-T-138 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: For proton therapy (PT) dose calculations, the Monte Carlo technique is the gold standard in terms of accuracy. At our institution a fast GPU-based Monte Carlo (GPUMC) dose engine has been developed and clinically deployed. We report on the use of octree compression (OC) of CT images for speeding up the GPUMC.
Methods: Using a fast GPU-based bottom-up OC algorithm, CT voxels with material-to-water stopping power ratio values within 3% are merged. HU values of voxels outside the external contour were overridden to air. Dose is scored only within selected structures (e.g. CTV) using the native CT voxel size, and computational times are recorded. Comparisons are made with the no-OC case.
Results: With OC, protons can be propagated with larger step sizes outside of structures requiring dose scoring. The compressed CT and CTV dose have been tested in three clinical cases, prostate, liver, and head-and-neck. When scoring CTV dose only, the time gain with the 3% OC voxelization is 11%, 12%, 30% for prostate, liver, and head-and-neck, respectively. Within the CTV, for the same sites, the dose difference distributions (fitted to Gaussians) between the native and the 3% OC voxelization cases, dnativeCT – dOC-CT, have been calculated as well. Difference values μ ± σ are: -7.5⋅10⁻⁵ ± 2.5⋅10⁻³ Gy for prostate, -2.9⋅10⁻⁶ ± 5⋅10⁻³ Gy for liver, and -1.06⋅10⁻⁴ ± 2⋅10⁻² Gy for head-and-neck.
Conclusion: To our knowledge this is the first attempt to introduce OC in GPUMC simulations. The OC was able to decrease the computational time while maintaining the same dosimetric accuracy in the CTV.
Keywords
Not Applicable / None Entered.
Taxonomy
TH- External Beam- Particle/high LET therapy: Proton therapy – computational dosimetry-Monte Carlo
Contact Email