Applying pytorch toolkit to plan optimization for circular cone based robotic radiotherapy

• DOI: 10.1186/s13014-022-02045-y
• Publication Date: 2022-04-20T15:03:33Z
• Authors: Bin Liang; Ran Wei; Jianghu Zhang; Yongbao Li; Tao Yang; Shouping Xu; Ke Zhang; Wenlong Xia; Bin Guo; Bo Liu; Fugen Zhou; Qiuwen Wu; Jianrong Dai
• Keywords: Solver

Abstract Background Robotic linac is ideally suited to deliver hypo-fractionated radiotherapy due its compact head and flexible positioning. The non-coplanar treatment space improves the delivery versatility but complexity also leads prolonged optimization time. Methods In this study, we attempted use deep learning (pytorch) framework for plan of circular cone based robotic radiotherapy. problem was topologized into a simple feedforward neural network, thus transformed network training. With transformation, pytorch toolkit with high-efficiency automatic differentiation (AD) gradient calculation used as solver. To improve efficiency, plans fewer nodes beams were sought. least absolute shrinkage selection operator ( lasso ) group employed address “sparsity” issue. Results AD-S (AD sparse) approach validated on 6 brain liver cancer cases results compared commercial MultiPlan (MLP) system. It found that achieved rapid dose fall-off satisfactory sparing organs at risk (OARs). Treatment efficiency improved by reduction in number (28%) (18%), monitor unit (MU, 24%), respectively. computational time shortened 47.3 s average. Conclusions summary, first attempt applying promising has potential be clinically.