GPU-accelerated Monte Carlo convolution/superposition implementation for dose calculation.

Area of Science:

• Medical Physics
• Computational Science
• Radiation Oncology

Background:

• Accurate dose calculation is critical in radiation treatment planning, especially with advanced therapy techniques.
• Current methods often involve a trade-off between speed (deterministic methods like convolution/superposition) and accuracy (Monte Carlo methods).
• The hybrid Monte Carlo convolution/superposition (MCCS) method combines strengths but requires optimization for clinical application.

Purpose of the Study:

• To enhance the performance of the MCCS dose calculation method.
• To develop a practical, day-to-day radiation therapy planning tool by implementing MCCS on a graphics processing unit (GPU).

Main Methods:

• Modified the MCCS algorithm to optimize for GPU hardware architecture.
• Implemented a GPU-based version of the MCCS algorithm.
• Compared the performance of the GPU implementation against a multithreaded CPU software implementation.

Main Results:

• Achieved a significant speedup of 6.7-11.4x for clinical cases using the GPU implementation.
• Demonstrated accuracy comparable to CPU implementation, with less than 2% statistical fluctuation.
• Validated the feasibility of GPU acceleration for MCCS dose calculations.

Conclusions:

• GPU implementation offers a feasible and cost-effective solution for accelerating dose calculations compared to clusters or FPGAs.
• This approach addresses the increasing computational demands for speed and accuracy in radiation therapy.
• The study also identifies and analyzes inherent limitations of using GPUs for Monte Carlo-type applications.