Recent developments and comprehensive evaluations of a GPU-based Monte Carlo package for proton therapy

Area of Science:

• Medical Physics
• Computational Biology
• Radiotherapy Physics

Background:

• Monte Carlo (MC) simulation is the most accurate method for proton therapy dose calculation.
• Existing MC tools may lack the speed required for clinical applications.
• Previous development of a GPU-based MC tool, gPMC, aimed to address this speed limitation.

Purpose of the Study:

• To report updates on the gPMC tool, focusing on accuracy, portability, and functionality.
• To introduce gPMC v2.0, developed in OpenCL for cross-platform compatibility.
• To present comprehensive tests validating the performance of gPMC v2.0.

Main Methods:

• Developed gPMC v2.0 using the OpenCL environment for enhanced portability.
• Refined physics models for nuclear interactions to improve calculation accuracy.
• Expanded scoring functions to include particle fluence, dose by particle type, and dose-averaged linear energy transfer (LETd).
• Implemented a multiple counter approach to improve scoring efficiency.

Main Results:

• gPMC v2.0 demonstrated improved accuracy in both water phantom and patient cases compared to gPMC v1.0.
• Substantially reduced dose discrepancies in a prostate cancer case against TOPAS, improving gamma test passing rate to 93.1%.
• Achieved average relative differences of 1.7% for LETd and within 2.3% for dose by different particle types.
• Simulated 10^8 protons in 8-17 seconds on a GPU with <1% statistical uncertainty.
• The multiple counter approach significantly reduced performance impact from reduced beam size.

Conclusions:

• gPMC v2.0 offers enhanced accuracy, portability, and efficiency for proton therapy dose calculations.
• The OpenCL development enables successful execution across various CPUs and GPUs.
• The tool's performance improvements make it a viable option for faster clinical dose calculations in proton therapy.