Fully Automated Treatment Planning Using the Pareto Optimal Projection Search (POPS) Algorithm.

Purpose/Objective(s) Radiation therapy treatment planning has traditionally been a time-consuming process with potentially high inter-planner variability. The Pareto optimal projection search (POPS) algorithm, which produces Pareto optimal and clinically acceptable plans, provides a fully automated framework for treatment planning without requiring any active planning. Materials/Methods The POPS algorithm searches the feasibility boundary for desirable plans, as defined by a scoring function. Overall, POPS has two main components—gradient-free search in the decision variable space and projection of decision variables to the Pareto front using the bisection method. As treatment plan evaluation is highly nuanced with many candidate scoring functions available in literature, we evaluate the performance of the POPS automated plans under the following two conditions: 1) when POPS performs plan scoring using a normalized equivalent uniform dose (NEUD) metric and 2) when POPS performs plan scoring using a scoring function (SF) based on dose-volume histogram (DVH) limits. Results Treatment plans were generated automatically using the POPS algorithm for a dataset of 21 prostate cases collected as part of clinical workflow. Dose conformity, dose homogeneity, and sparing of organs-at-risk are assessed using a Wilcoxon signed-rank test. In general, plans generated using the POPS algorithm were of high quality. Plans scored using the NEUD metric had improved conformity and comparable homogeneity as compared to plans scored using the SF score. Sparing of organs-at-risk for POPS generated plans was also comparable when using either of the scoring functions. Conclusion The POPS algorithm allows for fully automated treatment planning and produces plans that are pareto optimal and clinically acceptable. We anticipate that POPS will substantially improve treatment planning workflow.