ESTRO 2025 - Abstract Book

S3809

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

(Table 1). Trade-off was observed in organ-at-risks (OARs) sparring. Both CNN-Auto and DRL-Auto plans had lower total monitor units (MUs) compared with clinical plan, indicating improved modulation efficiency.

Conclusion: Both supervised and deep reinforcement learning approaches generate clinically acceptable treatment plans with comparable plan quality. The deep reinforcement learning-based framework provides a flexible and more effective solution, with the potential to further enhance clinical workflow efficiency.

Keywords: Deep learning, reinforcement learning

References: [1] Li, X., et al. "Commissioning of an Artificial Intelligence (AI) Tool for Automated Head and Neck Intensity Modulated Radiation Therapy (IMRT) Treatment Planning." International Journal of Radiation Oncology, Biology, Physics 114.3 (2022): S39. [2] Yang, D., et al. "Automated Treatment Planning with Deep Reinforcement Learning for Head-and-Neck Cancer Intensity Modulated Radiation Therapy." International Journal of Radiation Oncology, Biology, Physics 120.2 (2024): S64.

3016

Digital Poster Spinal cord toxicity following reirradiation: an NTCP model accounting for recovery Vitali Moiseenko 1 , Jimm Grimm 2 , Wolfgang Tome 3 , Gopal Subedi 2 , Rachel Grimm 3 , Michael Milano 4 , Parag Sanghvi 1 , Issam El Naqa 5 1 Radiation Medicine and Applied Sciences, University of California San Diego, La Jolla, USA. 2 Radiation Oncology, Wellstar Health System, Marietta, USA. 3 Radiation Oncology, Montefiore Medical Center, Bronx, USA. 4 Radiation Oncology, University of Rochester, Rochester, USA. 5 Radiation Oncology, H. Lee Moffitt Cancer Center & Research Institute, Tampa, USA Purpose/Objective: The human spinal cord is considered to be slow-responding and susceptible to late effects. For many patients, the spinal cord is a dose-limiting organ for reirradiation; evidence-based guidelines to establish tolerance doses are needed. We analyzed dose-time recovery models of published human spinal cord reirradiation data to establish dose-response with varying assumptions to describe recovery.