ESTRO 2025 - Abstract Book

S2851

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2025

considered significant. Additionally, 5 radiation oncologists blindly reviewed 20 plans (10 auto- and 10 manual) and assessed their clinical acceptability and preference for treatment.

Results: Compared to the manually generated clinical head and neck plans, all auto plans achieved PTV D 95% coverage and critical organs at risk sparing without statistically significant change in average global D max (107.4% for manual vs 107.5% for automated plans). The auto-planning solution provided reduced maximum doses to brainstem and spinal cord (average reductions of 3.6 ± 0.1 Gy and 2.1 ± 1.1 Gy, respectively, all p <0.001), reduced average mean doses to contralateral submandibular gland, ipsilateral parotid, oral cavity, cochleae, larynx, contralateral parotid (reductions of 4.1 ± 1.2 Gy, 3.9 ± 0.4 Gy, 2.5 ± 0.1 Gy, 2.4 ± 0.2 Gy, 2.0 ± 1.4 Gy, 1.5 ± 0.1 Gy, respectively, all p < 0.03) and reduced average maximum doses to mandible and lips (reductions of 2.9 ± 2.8 Gy and 2.3 ± 1.2 Gy, respectively, all p < 0.04). In the blinded review by physicians, out of 50 responses 94% considered auto-plans clinically acceptable versus 86% for manual plans. Overall, 7 auto-plans were preferred for treatment, 1 was deemed equivalent, while only 2 manual plans were preferred. Conclusion: The automated treatment planning script significantly improved plan quality for HN cancer patients by reducing important dosimetric indices to organs at risk while maintaining target coverage and dose homogeneity. Radiation oncologists appreciate reproducibility and efficiency of the auto-planning script in generating high-quality plans within a short timeframe. Digital Poster Knowledge-based autoplanning improves efficiency and plan quality for larynx stereotactic radiotherapy Yao Zhao 1 , Dong Joo Rhee 1 , Congjun Wang 1 , Tucker Netherton 1 , Sara Lynn Thrower 1 , Kelli McSpadden 2 , Xin Wang 1 , Anna Lee 3 , Amy Catherine Moreno 3 , David Rosenthal 3 , Jack Phan 3 , He Wang 1 1 Department of Radiation Physics, The University of Texas MD Anderson Cancer Center, Houston, USA. 2 Radiation Therapeutic Physics, The University of Texas MD Anderson Cancer Center, Houston, USA. 3 Radiation Oncology Department, The University of Texas MD Anderson Cancer Center, Houston, USA Purpose/Objective: To develop an automated workflow for larynx stereotactic body radiation therapy (SBRT) treatment planning by employing knowledge-based models to establish dose constraints for organs-at-risk (OARs) and implementing Python-based scripting in RayStation for automation. Material/Methods: A retrospective cohort of 26 larynx patients, each prescribed 42.5 Gy in 5 fractions, was analyzed to develop a linear regression model. This model incorporated parameters including the target's relative location to the larynx, target volume, OAR volume, and distance to the carotid arteries. The model performance was evaluated, indicating the linear model provided an adequate representation for predicting optimal dose constraints in the test cohort. The in-house developed auto-planning algorithm integrated these predicted dose constraints, combined with Python-based scripting in RayStation, to enable automated treatment planning. Treatment plans were generated with a consistent arc arrangement and included patient-specific dose constraints. The algorithm iteratively optimized the plans by adjusting objective functions and minimizing hotspots without user intervention.The end-to-end workflow was tested on an independent cohort of nine patients. Auto-generated treatment plans were compared to clinically approved plans using quality metrics, including target coverage, maximum dose, conformity index (ratio between ROI volume covered by Rx and total Rx volume), homogeneity index (D98%/D2%), mean dose and V20Gy for the cricoarytenoid and normal larynx, as well as the maximum dose for the ipsilateral carotid artery. Additionally, all generated plans Keywords: autoplanning, head and neck cancer, AI 3117