A GPU-based multi-criteria optimization algorithm for HDR brachytherapy

摘要: Currently in HDR brachytherapy planning, a manual fine-tuning of an objective function is necessary to obtain case-specific valid plans. This study intends facilitate this process by proposing patient-specific inverse planning algorithm for prostate brachytherapy: GPU-based multi-criteria optimization (gMCO). Two engines including simulated annealing (gSA) and quasi-Newton optimizer (gL-BFGS) were implemented compute multiple plans parallel. After evaluating the equivalence computation performance these two engines, one preferred engine was selected gMCO algorithm. Five hundred sixty-two previously treated cases divided into validation set (100) test (462). In set, number Pareto optimal achieve best plan quality determined compared with physician-approved clinical Over 462 cases, clinically 428 (92.6%) 461 (99.8%) The target V100 coverage greater than 95% 288 (62.3%) 414 (89.6%) mean time 9.4 s generate 1000 In conclusion, gL-BFGS able thousands SA equivalent treatment within short frame. Powered gL-BFGS, ultra-fast robust brachytherapy. A large-scale comparison against physician approved showed that could be improved significantly reduced proposed