KEY POINTS
- This technical validation study evaluated RadMonteCarlo, a pre-release, multithreaded CPU-based Monte Carlo engine integrated into Radformation's ClearCalc independent dose verification platform. The engine was commissioned for a pencil beam scanning proton system operating between 70 and 226 MeV, using integrated depth-dose curves, spot profiles, and absolute-output measurements. Four anonymized clinical plans, representing prostate, bladder, lung, and brain treatments, were recalculated using three different relative stopping-power (RSP) models, generating 12 plan–RSP comparisons against RayStation Monte Carlo calculations.
- Three HU-to-RSP conversion approaches were evaluated while keeping other simulation parameters constant. These included a density-derived empirical model, a Bethe–Bloch-based calculation, and a Schneider stoichiometric tissue model. RadMonteCarlo used a modified dose planning method (DPM) condensed-history proton transport algorithm, incorporating continuous energy loss, multiple Coulomb scattering, and selected nuclear interactions. Patient heterogeneity was represented through voxel-specific density and RSP within a virtual-water transport approximation, allowing the effect of stopping-power assignment to be isolated from differences in the beam source model.
- Commissioning demonstrated close agreement with the measurements used to construct the source model. Mean Bragg-peak fit quality was R² = 0.99977, with individual values ranging from 0.99855 to 0.99996. Recalculated distal R80 differed from measurements by +0.03 to +0.85 mm, with a mean difference of +0.39 ± 0.19 mm, meeting the study's 1-mm consistency criterion at all commissioned energies. However, these were the same measurement datasets used for parameter fitting, so they demonstrate internal commissioning consistency rather than independent experimental validation.
- All 12 clinical plan–RSP combinations met the prespecified 95% gamma passing criterion at 3%/3 mm. Full-dose-matrix passing rates ranged from 99.1% to 99.9%, while target-restricted rates ranged from 95.5% to 100.0%. These findings indicate consistent agreement with the reference treatment planning system under conventional evaluation criteria despite using different stopping-power models. The authors emphasize that 95% at 3%/3 mm was a study benchmark, not a universally validated proton-specific secondary calculation acceptance threshold.
- Stricter gamma evaluation exposed substantially greater differences between stopping-power models. At 2%/2 mm, full-matrix passing rates ranged from 91.2% to 97.9%, whereas target-restricted values ranged from only 75.8% to 97.3%. Within the same plan, the difference between the highest and lowest target passing rates reached 19.1 percentage points for bladder, 18.3 for brain, 14.1 for prostate, and 12.6 for lung. The 90% reference line shown for 2%/2 mm was descriptive rather than a formal acceptance criterion.
- Direct comparison of the three RSP models revealed clinically relevant localized dose and range sensitivity. Differences in target mean dose ranged from −1.15% to +1.20%, while D95 differences ranged from −1.11% to +1.96% and D98 from −1.06% to +2.23%. Mean distal R80 shifts across beam axes ranged from −0.3 to +1.9 mm, with the Schneider model producing the shortest range in all four evaluated plans. The 95th-percentile absolute voxelwise dose difference reached 4.1%, while pairwise target-restricted 2%/2 mm gamma passing rates ranged from 84.8% to 99.8%. These differences depended on anatomy and beam paths rather than demonstrating that any one RSP model was physically superior.
- Clinical-default calculations were substantially faster than elevated-statistics research simulations. Using a cloud-based compute service and the default history sampling setting, the four plans were recalculated in 36–98 seconds, or approximately 47–113 seconds including data transfer, with estimated statistical uncertainty of 1.3–1.8% in high-dose regions. Publication-level calculations, using 30-fold more simulated histories, required 33 minutes to 8.3 hours locally and achieved approximately 0.3% uncertainty. The study was limited to one proton system, one clinical plan per anatomical site, and no independent measurement of patient-specific stopping power. All authors were employees of Radformation, the developer of the evaluated software.
CLINICAL TAKEAWAY
Independent Monte Carlo recalculation can provide clinically practical verification of proton treatment plans, but conventional gamma agreement may conceal localized differences arising from HU-to-RSP calibration. This study supports combining global gamma analysis with target-specific dose metrics, distal range evaluation, and review of beam-path heterogeneity. The findings are limited to four clinical plans and do not establish which stopping-power model is most accurate or validate universal proton QA acceptance criteria.