Purpose: In proton therapy dose calculation, conventional Monte Carlo (MC) simulation may not take full advantage of GPUs because the random secondary-particle-generating processes result in unbalanced workload for different threads of GPUs. Therefore, we propose a novel concept of virtual particle (VP) in MC (VPMC) to avoid simulating secondary particles with clinically acceptable accuracy in proton therapy dose calculation.
Methods: We interpreted the track histories from a conventional MC code, MCsquare, in a new way (Figure 1). The dose of one primary proton and its secondaries, named realistic particles, can be regarded as if it is deposited by multiple VPs, which results in the same dose distribution as realistic particles. Each VP corresponds to one realistic particle. However, VPs start at the same starting position of the primary proton. After converting the histories of realistic particles into histories of VPs, we can induce a probability model to describe the behaviors of VPs. The model includes a continuing-slowing-down-approximation (CSDA) model and a large angle event model corresponding to nuclear interactions for VPs. In this study, to simplify the calculation, we ignored neutrons and gamma rays, locally deposited the dose of electrons, heavy ions, and nuclear fragments, and converted the tracks of deuterons into tracks of protons. Finally the dose distribution from realistic particles was calculated by simulating only VPs in VPMC, which was then carefully benchmarked with the one calculated by MCsquare.
Results: The curves of integrated-depth dose, percentage-depth dose, iso-dose contours, and lateral-dose profiles all matched very well (Figure2). The 3D Gamma passing rate with a threshold of 1%/1mm/2%/2mm between VPMC and MCsquare is 97.8%/99.4% in water phantoms and 96.5%/99.4% in inhomogeneous phantoms, respectively. The calculation is highly efficient (1.3×10^7 particles per second with 6 Nvidia T4 GPUs).
Conclusion: VPMC achieved high accuracy and efficiency in proton therapy dose calculation.
Not Applicable / None Entered.