A phase space discretization model for efficient particle source sampling in boron neutron capture therapy

Efficient source sampling is important for Monte Carlo dose calculation in accelerator-based boron neutron capture therapy (AB-BNCT). This study presents a Phase Space Discretization Model (PSDM) that represents the weighted joint distribution of five source variables using a multidimensional histogram and a lookup table of occupied cells. Uniform cell selection followed by continuous sampling within the selected cell is combined with statistical weights to reproduce the discretized source distribution. For the LZU-AB-BNCT source, the stored representation was reduced from a 20.6 GB phase space file (PSF) to a 0.3 GB lookup table. In the reported source-operation benchmark for 2.2 × 108 histories, file I/O and source initialization/sampling took 92.73 min for the implemented PSF workflow and 0.15 min for PSDM, a ratio of approximately 618. This ratio concerns source operations and does not represent the speedup of the complete dose calculation or the isolated benefit of sparse storage. Comparisons of phase space distributions, depth-dose curves, and dose-volume histograms showed agreement for the tested configuration. A single glioblastoma patient voxel model yielded a total equivalent-dose gamma pass rate of 94.85% at 1%/1 mm, providing a demonstration of dosimetric feasibility. PSDM retains joint cell probabilities at the selected grid resolution; subcell structure, finite-source uncertainty, and downstream transport uncertainty remain. The results support the use of this compact source representation in the evaluated workflow, while grid sensitivity, statistical efficiency, and performance across additional treatment configurations require further assessment.