Purpose: We previously developed an Intelligent Automatic Treatment Planning framework with a virtual treatment planner (VTP), an artificial intelligence robot, operating a treatment planning system (TPS). Using deep reinforcement learning guided by human knowledge, we trained the VTP to autonomously make decisions to adjust relevant parameters in treatment plan optimization, similar to a human planner, to generate high-quality plans for prostate cancer Stereotactic body radiation therapy (SBRT). This study implements VTP in our clinic and evaluates its performance by comparing plans with human-generated plans.
Methods: We built an interface between VTP and the treatment planning interface of Eclipse TPS via Eclipse scripting API. Similar to a human planner, VTP observes dose-volume histograms of relevant structures and decides how to adjust dosimetric constraints of each structure in the Eclipse treatment planning interface, including their doses, volumes, and weighting factors. This process continues until a high-quality plan is achieved. We evaluated VTP’s performance in the case of the 2016 American Association of Medical Dosimetrist Plan Competition and compared it with human-generated plans submitted to the competition. Using the same Proknow scoring system in the competition, we evaluated the quality of plans for 10 prostate SBRT cases treated at our institution. We compared with plans generated by an experienced dosimetrist (3-year experience).
Results: In the planning competition case, VTP achieved a score of 141.8/150.0, ranking the third among all human planers who participated in the competition (mean 132.2 and median 134.6). For the 10 clinical cases, VTP achieved 128.3±4.3, slightly higher than plans generated by human planner 128.0±4.5.
Conclusion: VTP can operate a commercial TPS for autonomous human-like treatment planning in prostate cancer SBRT. Its performance indicates the potential of the Intelligent Automatic Treatment Planning framework in terms of providing planning expertise.
Funding Support, Disclosures, and Conflict of Interest: NIH R01CA237269, R01CA254377
Treatment Planning, Stereotactic Radiosurgery
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation