Preliminary Monte Carlo Benchmarking of the First U.S. Carbon TPS: Range, Voxel-Wise Differences, and Microdosimetry
Poster
aapm2026-carbon-de-validation-poster.pdf Hosted by AAPM · 0.7 MBAbstract
Purpose
To present preliminary validation results for the first carbon treatment planning system (TPS) in the United States and to benchmark its GPU Monte Carlo dose engine using depth-resolved dose and microdosimetric metrics.
Methods
Pristine carbon-ion beams at 100, 290, and 430 MeV per nucleon and four spread-out Bragg peaks (SOBPs) were simulated in homogeneous 30×30×50 cm³ phantoms composed of water, soft-tissue elemental surrogates (H, C, N, O), calcium, and titanium. Dose was scored on 1 mm isotropic voxels. Independent reference calculations were performed with OpenTOPAS/Geant4 v11.3.0 using the Light-Ion Quantum Molecular Dynamics model. Agreement metrics included distal 80% range (R80), integral dose, and depth-resolved voxel-wise energy-deposition differences normalized to the total reference energy in the slab at Bragg Peak. For SOBPs, microdosimetric spectra were evaluated at six characteristic depths, with derived dose-averaged LET (LETd) and dose-mean lineal energy.
Results
R80 differences were within water-equivalent 0.7 mm while integral dose differences show larger differences from -0.07% to -4.33% across all materials. Depth-normalized voxel-wise analysis showed that most voxels were within 1%, while localized hotspots up to approximately 2% appeared in distal high-gradient regions and higher-Z media. For comparison, a 6 MeV photon beam in a water phantom with 1 × 1, 10 × 10, and 30 × 30 cm² field sizes yielded a maximum voxel-wise difference of 0.34%. For carbon SOBPs, plateau dose agreement was within 1.5%, while microdosimetric spectra showed high agreement with less than 1% error across all depths.
Conclusion
These preliminary results support commissioning of the first U.S. carbon TPS and demonstrate that depth-resolved voxel-wise and microdosimetric analyses provide sensitivity beyond conventional metrics.