ESTRO 2025 - Abstract Book

S3398

Physics - Machine learning models and clinical applications

ESTRO 2025

Conclusion: With the cyclic LSTM longer prediction horizons can be achieved with low prediction error. Furthermore, a low prediction error could already be achieved with a training dataset comprising only 30 patients, underlining the data efficiency of this network.

Keywords: Motion prediction, LSTM, Breathing motion

References: [1] Lin et al. doi: 10.1088/1361-6560/ab13fa [2] Renner et al. doi: 10.1016/j.phro.2024.100594.

2463

Mini-Oral Full dysphagia toxicity score trajectory prediction using deep learning; a multi-label and sequential approach Luuk van der Hoek, Daniel C. MacRae, Suzanne P.M. de Vette, Robert van der Wal, Hendrike Neh, Nanna M. Sijtsema, Johannes A. Langendijk, Peter M.A. van Ooijen, Lisanne V. van Dijk Radiotherapy, University Medical Center Groningen (UMCG), Groningen, Netherlands Purpose/Objective: Dysphagia significantly affects quality of life of head and neck cancer (HNC) patients treated with radiotherapy. Predictive modelling for radiation-induced dysphagia can affect clinical decision making, such as guiding dose optimization and selecting between proton and photon irradiation. To date, prediction models focused on predicting one post-treatment time-point, failing to capture the full toxicity burden experienced by patients. This study explored different modelling approaches to predict the full toxicity score trajectory of dysphagia in HNC patients, providing new avenues for patient-specific care. We compared four modelling approaches: 1) logistic regression, 2) per time-point deep learning, 3) multi-label deep learning, and 4) sequential deep learning. Material/Methods: This study included 1,090 HNC patients who received definitive (chemo)radiotherapy between 2007 and 2021 at a single medical centre. The endpoints were physician-rated moderate-to- severe dysphagia (CTCAEv4.0 grade≥2). Dysphagia toxicity scores were prospectively collected across 14 time-points: weekly during treatment (6 time points: week 2-7), and 6 weeks, 6, 12, 18, 24, 36, 60 months after treatment.