ESTRO 2025 - Abstract Book

S3419

Physics - Machine learning models and clinical applications

ESTRO 2025

Results: Over 30,000 samples (input/output of models) were considered in the liver, bone, and lung datasets originating from 40, 70, and 90 treatment plans, respectively. Mean absolute percentage error differences between the predicted and clinical MW per CP and MU per beam were smaller than 7% in all cases. The 3%/2mm gamma passing rates were mean±SD=98.7±2.9% and 98.3±3.6% for the single and multi-site model, respectively and no statistical difference was observed. Target coverage (PTV D95% and D99%) and organs at risk dose metrics measured in AI RTPlans were within ±2% of the clinical value in all cases (example of lung doses in Fig 2). The PTV D2% was within ±4% of the clinical value in all AI-RTPlans. The differences in dose conformity, in terms of CI100 and CI50, were respectively smaller than 7% and 10%.

Conclusion: Combining different anatomical locations for deep-learning prediction of the MU per CP is an accurate strategy in the automation of single-site radiation therapy treatment planning of oligometastases.

Keywords: AI, deep learning, oligometastatic cancer

References: [1] Guo C, Huang P, Li Y, Dai J. Accurate method for evaluating the duration of the entire radiotherapy process. J Appl Clin Med Phys 2020;21:252 – 8. [2] Bauman GS, Corkum MT, Fakir H, Nguyen TK, Palma DA. Ablative radiation therapy to restrain everything safely treatable (ARREST): study protocol for a phase I trial treating polymetastatic cancer with stereotactic radiotherapy. BMC Cancer 2021;21:405. [3] Palacios MA, Verheijen S, Schneiders FL, Bohoudi O, Slotman BJ, Lagerwaard FJ, et al. Same-day consultation, simulation and lung Stereotactic Ablative Radiotherapy delivery on a Magnetic Resonance-linac. Phys Imaging Radiat Oncol 2022;24:76 – 81.