Room: AAPM ePoster Library
Purpose: A critical step to ensuring automated planning systems generate clinically acceptable plans is a quantitative comparison to previous planning practices. The recently-released Varian Ethos platform generates treatment plans using an Intelligent Optimization Engine (IOE) without traditional user-navigated inverse optimization. The purpose of this study was to evaluate the quality of Ethos-generated plans compared to plans generated by our clinically-validated external beam knowledge-based planning routines for head-and-neck (H&N), gynecologic (GYN), and prostate cancers.
Methods: The IOE seeks to maximize the quality of the final dose distribution based on a prioritized list of clinical goals specified by the user. It automatically creates internal optimization structures and modifies objectives during optimization. For this study, a subset of patients treated retrospectively on our Halcyon (same beam model as Ethos) were identified and replanned using Ethos. These included 5 H&N, 5 GYN, and 8 prostate cases. All plans were created using 12-field IMRT. For each treatment site, our clinical standards were provided to Ethos as goals for plan generation. Resultant automated Ethos plans were exported to Eclipse and separately planned using our clinically-validated knowledge-based planning models. DVH metrics were extracted for each and compared using Wilcoxon signed-rank test with Bonferroni correction.
Results: Compared to our clinical plans, GYN Ethos plans had higher PTV doses while OAR sparing showed improvements for bladder and rectum in the medium-low dose region. Bowel and bone marrow sparing was unchanged. H&N Ethos plans also showed small increases in PTV dose while maintaining equivalent OAR dose sparing. Prostate PTV and OAR metrics improved or stayed consistent with Ethos planning. None of the compared metrics were significantly different on statistical tests.
Conclusion: Our initial results show that the fully-automated Ethos IOE generated clinically-equivalent plans to existing knowledge-based models that have been in routine clinical use and validated across hundreds of patients.
Funding Support, Disclosures, and Conflict of Interest: Dr. Ray has a lab services agreement with Varian Medical Systems. Dr. Moore reports income for personal consulting and speakers honoraria from Varian Medical Systems. Dr. Bojechko, Receives grant funding from Varian Medical Systems. This work was supported by an internal grant from the Center for Precision Radiation Medicine.
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation