ESTRO 2025 - Abstract Book

S1433

Clinical - Lung

ESTRO 2025

study is to develop and evaluate an artificial intelligence module to automatic generate clinically acceptable treatment plans.

Material/Methods: An auto-planning module was developed by RaySearch based on a contextual Atlas Regression Forest. Treatment plans of five lung SBRT patients with single target that were treated using fractionation scheme of 3x18 Gy was delivered to RaySearch as configuration dataset to adapt the module. This adapted module was used to generate plans for another 18 cases. The results were analyzed and the module was further adapted until satisfactory results were obtained. All plans were generated in Raystation using 2 6MV VMAT arcs. The dose-volume histogram end point of the auto-plans from the final accepted module were compared to those of the clinical plans. All auto-plans were also reviewed by the radiation oncologist to ensure their clinical acceptability. Results: Target coverage for both PTV and ITV were comparable between auto-plans and clinical plans (figure 1a). V 140% for auto-plans was slightly higher than the clinical plans, 0.3% vs 0%, respectively, resulting in slightly lower conformity index in auto-plans compared to clinical plans, 0.80 vs 0.84, respectively. In the first few modules, auto-planning struggled to meet several lung constraints, but after a few adaptations, the OARs doses were comparable with those of the clinical plans (figure 1b). The differences were all within the error margin and may be clinically negligible. Even though the differences may not be significant, in 17 of the OARs parameters analyzed, 15 were found to be lower in auto-plans and 2 in the clinical plans. As for the error margin, 10 were found to be smaller in the auto-plans and 7 in the clinical plans. These findings indicate that auto-plans are slightly better in plan quality and consistency.