Development of a GPU-Accelerated Continuous-Energy Monte Carlo Method for BNCT Dose Calculation
Abstract
Purpose
Accurate and efficient dose calculation remains a major challenge in boron neutron capture therapy (BNCT) due to the complexity of neutron transport and the contribution of secondary particles. Conventional Monte Carlo (MC) methods are often too computationally demanding for routine clinical use. This study aims to develop and validate a GPU-accelerated, continuous-energy MC method for BNCT dose calculation with improved computational efficiency while maintaining accuracy.
Methods
A GPU-accelerated continuous-energy MC neutron transport code, ARCHER-Neutron, was developed for BNCT dose calculation. Neutron transport physics and nuclear data were implemented based on Geant4 11.4.0 (physics list: QGSP_BIC_AllHPT), including elastic scattering, inelastic scattering, and capture, while thermal neutron scattering below 4 eV was modeled using the S(α,β) formalism. To improve GPU efficiency, neutron cross-section data were reorganized onto a unified energy grid for fast lookup during particle transport. A phase-space neutron source was used to represent realistic BNCT beam conditions. Secondary photons and electrons were simulated using the GPU-based MC code ARCHER. The method was validated against Geant4 in water phantom and voxelized patient phantom by comparing neutron flux and biologically weighted dose distributions.
Results
The proposed method showed good agreement with Geant4. In the water phantom with B-10 (30ppm), neutron flux distributions exhibited a maximum relative error below 2%. In the patient phantom, a 1 mm/1% gamma passing rate exceeding 99% is achieved. Regarding computational performance, the proposed method achieved up to a 30-78 speedup compared with Geant4 executed on a 128-threads CPU.
Conclusion
This study demonstrates that GPU-accelerated MC approach can simultaneously achieve high accuracy and high computational efficiency for BNCT dose calculations. By enabling fast and reliable dose simulations in patient phantoms, the proposed method shows strong potential for integration into BNCT treatment planning systems and for supporting clinically practical dose evaluation.