Fast Dose Calculations in Radiation Therapy with GPUs

Abstract

Dose calculations are central to planning procedures in radiation therapy. For advanced delivery techniques such as IMRT and arc therapy, dose typically is computed multiple times in optimization strategies. This often leads to computational burden and possible delays, that require for solution either significant computing resources or approximations that may compromise the accuracy of the calculations. In order to obtain accurate solutions within clinically compatible time-frames, we have investigated the use of graphics material for the calculation of dose in radiation therapy. The massively parallel architecture of Graphics Processing Units (GPUs) was exploited to accelerate dose calculation algorithms, which are inherently compatible with parallel computing. More specifically, we have implemented on graphics material a convolution-superposition (CS) dose calculation algorithm for external beam radiation therapy and a modified version of a broadly used protocol for brachytherapy. In both cases, the algorithms rely on raytracing through voxel space to account for phantom composition and geometry. Implementation strategies are described, along with strengths and limitations of GPUs as general-purpose computation engines. Acceleration factors of up to 17x were obtained for the CS dose calculation and of up to 25x for the brachytherapy dosimetry protocol, compared to execution on a Central Processing Unit (CPU). These results let envision the development of more accurate dose calculation algorithms compatible with clinical timescales.