2160 - A Knowledge-Based-Planning Model for Clinical Trial QA of Radiation Therapy Plans for High-Grade Glioma

Purpose/Objective(s): We have built and validated a knowledge-based-planning (KBP) model for radiation therapy (RT) planning of high-grade glioma that can be used for automated clinical trial QA of radiotherapy treatment plans. To demonstrate its utility, we retrospectively audited RT plans from other institutions.

Materials/Methods: Cases treated outside of our institution but within our health network were replanned using our KBP model. Thirty-four challenging clinical plans were selected to include overlap between planning target volumes (PTVs) and organs at risk (OARs). Clinical plans were optimized for two or three arc volumetric modulated arc therapy with the isocenter placed at the discretion of the planner, using 6 MV photons and either full or partial arc geometries. All KBP plans used two full arcs with the isocenter placed at the PTV center of mass and 6 MV photons. KBP-based optimization ran with no planner input to achieve the final dose distribution. Consensus dose objectives for targets and OARs were used to compare clinical and KBP plans. Plans were classified per objective as (1) Clinical plan was superior, (2) Clinical plan was within 5% of KBP, or (3) Clinical plan could be improved. The clinical plan was characterized as superior for an objective if the metric was > 5% better than in the KBP plan, or if the clinical plan met the objective but the KBP plan did not. Possible clinical plan improvement was indicated for an objective if the metric was > 5% better in the KBP plan, or if the KBP plan met the objective but the clinical plan did not. Lateral structures were combined for a single assessment.

Results: Table 1 summarizes the classifications for selected OAR objectives. Clinical plans could have been improved between 24% (Brainstem D_max) and 68% (Optic Nerve D_max) of the time, while the clinical plan was superior between 9% (Optic Nerves and Chiasm D_max, and Optics PRV V_54Gy) and 51% (Eyes D_max) of the time. The patterns of potential improvement and superiority of clinical plans differed between institutions. Improvement for PTV coverage (V_95%) was possible in 18% of clinical plans, while only 1 case was identified where the clinical plan was superior to the KBP plan. In 31 of 34 cases plan coverage was either met (9) or not met (22) in both plans, reflecting the challenging nature of the cases.

Conclusion: Our KBP model was evaluated as a tool for clinical trial QA. It successfully identified opportunities for plan improvement in two-thirds of cases.