SU-E-T-476: GPU-Based Monte Carlo Radiotherapy Dose Calculation Using Phase- Space Sources

Area of Science:

• Medical Physics
• Computational Science
• Radiotherapy Physics

Background:

• GPU-based Monte Carlo (MC) algorithms accelerate dose calculations but can suffer from inefficiency due to thread divergence when processing phase-space files sequentially.
• Optimizing the utilization of phase-space data is crucial for improving the speed of GPU-accelerated MC simulations in radiotherapy.

Purpose of the Study:

• To develop and implement an efficient strategy for integrating phase-space source models into the GPU-based MC dose calculation engine, gDPM.
• To enhance the performance of GPU-based MC simulations by optimizing phase-space file utilization.

Main Methods:

• Developed a novel method to group particles from phase-space files into phase-space-lets (PSLs) based on type, energy, and spatial location.
• Implemented PSL utilization within the gDPM engine to process only particles within the secondary collimator field opening.
• Validated the gDPM PSL implementation by comparing its results with DOSXYZnrc using a BEAMnrc phase-space source.

Main Results:

• The gDPM PSL implementation demonstrated agreement within 2% compared to DOSXYZnrc for various field sizes in a water phantom.
• Significantly reduced calculation times: 54 GPU seconds for gDPM vs. 147 CPU hours for DOSXYZnrc (650 million histories, 10x10 cm field).
• Achieved a 99.54% gamma pass rate (3 mm/3% criteria) for a realistic 7-field IMRT tongue treatment plan, with gDPM calculation times comparable to DOSXYZnrc.

Conclusions:

• An efficient method for utilizing phase-space sources in GPU-based MC dose calculations has been successfully developed and implemented in gDPM.
• The PSL strategy enhances computational efficiency without compromising dose calculation accuracy.
• This advancement offers a faster and effective approach for MC-based dose calculations in radiotherapy planning.