Development of a GPU-Based Dose Calculation Engine for MRI-Guided Proton Therapy πŸ“

Author: Yaowen Cao, Haonian Gong, Yue Gu, Meiqi Liu, Hsiao-Ming Lu, Yuxiang Wang, Yidong Yang, Xiaogang Yuan πŸ‘¨β€πŸ”¬

Affiliation: Hefei Ion Medical Center, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Department of Engineering and Applied Physics, University of Science and Technology of China 🌍

Abstract:

Purpose: Magnetic resonance imaging (MRI)-guided proton therapy is an innovative technology that combines superior soft tissue imaging capabilities of MRI with the high dose conformity of proton therapy. For clinical implementation of MRI-guided proton therapy (MRPT), a fast and accurate dose calculation engine that accounts for the effects of MRI’s heterogeneous magnetic field on dose delivery is necessary. This study aims to develop a GPU-accelerated proton and ion dose engine (PRIDE) that enables fast and accurate dose computation in MRPT for practical clinical configurations.
Methods: Pencil beam parallelization and voxel parallelization were integrated for high speed dose calculation. The region of interest for dose superposition was confined to the scattering-related region to further improve efficiency. A 3D magnetic field map of a 1.5 T split-bore MRI system was simulated in COMSOL Multiphysics and employed for PRIDE validation. The phase space source was positioned 1.5 m from the isocenter to ensure that proton beams traverse both the fringe field and the imaging field. PRIDE was validated against Monte Carlo (MC) simulations using two practical plans (a water phantom plan and a prostate tumor plan) under two possible hardware configurations: beam parallel (inline) to the magnetic field and beam perpendicular to the magnetic field.
Results: GPU-based PRIDE demonstrated excellent accuracy and efficiency in both beam configurations. The 2%/2 mm gamma passing rates at 10% threshold exceeded 98.7% for the two evaluated plans. The computation time for each plan was less than 4 s on an NVIDA GeForce RTX 3060 GPU.
Conclusion: GPU-based PRIDE can perform fast and accurate dose computations in the heterogeneous magnetic field of a practical split-bore MRI scanner in both the inline and perpendicular orientation. Its excellent performance and versatility make it suitable for future clinical implementation.

Back to List