GPU-accelerated scanning path optimization in particle cancer therapy

Abstract

When using the beam scanning method for particle beam therapy, the target volume is divided into many iso-energy slices and is irradiated slice by slice. Each slice may comprise thousands of discrete scanning beam positions. An optimized scanning path can decrease the transit dose and may bypass important organs. The minimization of the scanning path length can be considered as a variation of the traveling salesman problem; the simulated annealing algorithm is adopted to solve this problem. The initial scanning path is assumed as a simple zigzag path; subsequently, random searches for accepted new paths are performed through cost evaluation and criteria-based judging. To reduce the optimization time of a given slice, random searches are parallelized by employing thousands of threads. The simultaneous optimization of multiple slices is realized by using many thread blocks of general-purpose computing on graphics processing units hardware. Running on a computer with an Intel i7-4790 CPU and NVIDIA K2200 GPU, our new method required only 1.3 s to obtain optimized scanning paths with a total of 40 slices in typically studied cases. The procedure and optimization results of this new method are presented in this work.