Abstract: The Monte Carlo (MC) code is essential for accurate radiotherapy dose calculation. In proton therapy, the accuracy of dose calculation algorithms has a more significant impact than photon therapy due to the depth dose characteristics of proton particles. MC code requires considerable computation cost to achieve accurate results. This was a major hurdle in implementing MC code in routine clinical practice. There have been efforts to develop simplified physics models and hardware-based acceleration algorithms to minimize computational cost without degrading accuracy. Among the algorithms, accelerating computation using the general-purpose graphic processing unit (GPGPU) achieved a reasonable computation burden with clinically acceptable accuracy. Contrary to the central processing unit (CPU), GPGPU has limited memory and is not expandable. This is a hurdle to score quantities with large dimensions, which requires large memory. In this study, we developed a novel GPGPU-based MC code with a hash table, one of the key-value pair data structures, to fully utilize limited memory attached to GPGPU. With the hash table, only voxels interacting with particles occupy the memory, and we can search over the data efficiently to determine the address of data. The developed code was validated with MC code widely used in clinical settings, TOPAS. We also compared the dose calculation results for prostate, liver, and head and neck (H&N) cases using different MC codes, including one accompanied by a commercial treatment planning system (TPS), RayStation (RaySearch, Stockholm, Sweden). The gamma evaluation was done with 2mm/2% criteria to compare the results. Developed code achieved more than 99% of gamma pass rate compared to TOPAS and more than 95% of gamma pass rate compared to RayStation MC. Using the developed code, it took less than one minute per beam field to calculate patients' dose distribution, and results showed good agreement with the other MC codes.

About the speaker: I’ve got my Ph. D. in Nuclear and Quantum Engineering at Korea Advanced Institute of Science and Technology (KAIST). During my Ph. D., I studied low-dose CT imaging and automating IMRT planning with deep learning methods. I’m currently working as a postdoctoral researcher in Radiation Oncology at Massachusetts General Hospital. Currently, I’m working on GPGPU-based fast Monte Carlo algorithms, and CBCT scatter correction for adaptive proton therapy and imaging applications of TOPAS.