Multi-Institutional Evaluation of a Universal GPU-Accelerated Monte Carlo Virtual-Source Model for Proton Therapy Commissioning and QA
Abstract
Purpose
Although the number of proton centers continues to rise worldwide, rapid methods for commissioning and dose verification are still lacking. To evaluate the performance of a GPU-accelerated Monte Carlo (MC) dose computational framework for patient-specific plan QA at 4 participating centers using clinical data.
Methods
Virtual-source models of the treatment head of multiple proton delivery system (IBA ProteusOne, Varian ProBeam, Siemens, Mevion S250i) were commissioned and integrated into a GPU-paralleled MC simulation platform, ARCHER which contains detailed proton radiation physics models to support proton dose calculations on patient CT images and comparisons with dose distributions from commercial treatment planning system (TPS) by gamma index analysis. A total of 427 clinical treatment cases (comprising of 170 head and neck (HN),159 thoracic, and 98 abdominal cases) were selected from participating centers. All simulations were performed on a workstation equipped with one NVIDIA RTX 3090 GPU to evaluate computational efficiency.
Results
Mean gamma pass rate (GPR) were 99.31%, 99.54% and 99.53% for HN, abdominal and thoracic cases, respectively. ARCHER identified several lung cases that were calculated by pencil beam algorithm for Varian ProBeam system with poor GPR of ~ 80%. Average calculation time was 19 and 0.2 seconds with and without AI-based MC denoising tool.
Conclusion
Results from 427 clinical patient cases at multiple centers using different proton treatment platforms show that the accuracy and speed of ARCHER suitable for clinical secondary dose verification. On-going development continues to support a wider range of proton delivery platforms.