Plan-Delivery-Time-Constrained Inverse Optimization for High-Throughput PBS Proton RT

Abstract

Proton RT remains a limited resource that can only accommodate small number of cancer patients. Given that most modern proton centers use the multi-room shared-beam layout to parallelize patient setup, the major throughput-limiting factor is plan delivery time (primarily dose delivery time and energy switching time). This project will develop novel plan-delivery-time-constrained inverse optimization algorithms to achieve the significant reduction of plan delivery time for high-throughput pencil beam scanning (PBS) proton RT. The PBS dose rate is linearly proportional to the minimum-monitor-unit (MU) constraint. Thus, the increase of MU threshold can increase dose rate and therefore reduce dose delivery time. However, the room to increase this threshold is very limited with a single fixed MU threshold for all PBS spots as is assumed by current methods, since the plan quality is often significantly deteriorated before reaching any meaningful reduction of dose delivery time. To allow a significant reduction of dose delivery time while ensuring the plan quality, we propose to use variable energy-dependent MU thresholds to relax the tradeoff between dose delivery time and plan quality. On the other hand, although energy switching time is the secondary factor for total plan delivery time, it becomes relatively more important as dose delivery time is significantly reduced. Thus, energy switching time is modeled via group sparsity. New optimization algorithms via iterative convex relaxation are developed to solve these non-convex optimization problems. In addition, our methods can provide the efficiency-quality Pareto surface via multi-criteria optimization regarding the tradeoff between delivery efficiency and plan quality. With variable energy-dependent MU thresholds, our new method achieved 46.7%, 32.6%, and 33.3% reduction of dose delivery time respectively for prostate, lung and brain cases. Compared with plans optimized with a single MU threshold that corresponded to a 26.7% reduction of dose delivery time, our new method significantly improved plan quality in terms of both CTV coverage and OAR sparing. Moreover, the dose delivery time was further reduced to 25.7%, 34.9%, and 41.3% respectively, accepting a slightly degraded plan quality that still met all clinical planning constraints. On the other hand, our group-sparsity-regularized method reduced number of energy layers (or equivalently energy switching time) to 78%, 76%, and 61% respectively for prostate, lung and HN cases, while comparable plan quality was preserved. Moreover, the energy switching time was further reduced to 54%, 57%, and 37% respectively, accepting a slightly degraded plan quality that still met all clinical planning constraints. Novel plan-delivery-time-constrained optimization methods have been developed for high-throughput PBS proton RT.