Development of an adjoint Monte Carlo method for clinical prostate radiation treatment

Adjoint Monte Carlo methods were originally developed for optimizing nuclear reactor designs, but have recently shown to be useful in radiation treatment planning. However, the theoretical framework of Adjoint Monte Carlo (AMC) method was not designed and was never demonstrated with realistic patient anatomy for clinical procedures. In this project, we developed and extended the AMC framework in the MCNP code to a 3D anatomical model called VIP-Man. A case of prostate tumor treatment was considered in implementing and testing the AMC method. This thesis evaluated the efficiency of the AMC method and addressed several issues on implementing the method into the MCNP code. We compared the timing of adjoint and a forward Monte Carlo calculation. By exploring a tally feature (the FT card) of MCNP, we developed a source-sampling scheme for multiple regions of interest in a single AMC simulation in the MCNP code. We further implemented a voxel-based source-sampling scheme, which supplies necessary information for treatment plan optimization. Two methods were developed and implemented in this project to speedup Monte Carlo simulations: a mesh tally speedup and a variance reduction technique (VRT) using importance functions from the adjoint Monte Carlo calculations. The comparisons showed that the adjoint MC run was about 9 times faster than the forward MC run. The source energy biasing method reduced the computation time by 26.7% in the forward MC calculation for prostate. The time reduction was 24.8% for the urinary bladder and 13.6% for the rectum. This study is one of the first and the only study that demonstrates the AMC in a clinically meaningful patient anatomy. The tools developed in this study have bridged the cap between the theory that was originally developed by the reactor physics community and clinically meaningful procedure such as 3D-CRT of prostate cancer.