ESTRO 2025 - Abstract Book

S3794

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

2567

Poster Discussion A Probabilistic Model to Predict Lymph Node Metastasis in Oral Cavity Squamous Cell Carcinoma Based on Multi-Institutional Data Roman Ludwig 1,2 , Yoel S Perez Haas 1,2 , Panagiotis Balermpas 1 , Adrian Schubert 3,4,5 , Dorothea Barbatei 6 , Laurence Bauwens 6 , Olgun Elicin 3 , Matthias Dettmer 7,8 , Roland Giger 4 , Vincent Grégoire 6 , Jan Unkelbach 1,2 1 Radiation Oncology, University Hospital Zurich, Zurich, Switzerland. 2 Physics, University of Zurich, Zurich, Switzerland. 3 Radiation Oncology, Inselspital, Bern University Hospital, Bern, Switzerland. 4 Head and Neck Surgery, Inselspital, Bern University Hospital, Bern, Switzerland. 5 Head and Neck Surgery, Réseau Hospitalier Neuchâtelois, Neuchatel, Switzerland. 6 Radiation Oncology, Centre Léon Bérard, Lyon, France. 7 Pathology, Klinikum Stuttgart, Stuttgart, Germany. 8 Tissue Medicine and Pathology, Bern University Hospital, Bern, Switzerland Purpose/Objective: Current guidelines for defining the elective clinical target volume (CTV-N) in oral cavity squamous cell carcinoma (OCSCC) often recommend extensive bilateral irradiation based on the per-level prevalence of lymph node involvement [1]. These recommendations do not account for individual patient's diagnoses and spread patterns, potentially causing unnecessary treatment toxicity. We adapted a previously developed statistical model for oropharyngeal SCC [2] to a multi-institutional dataset of OCSCC patients to estimate patient-specific risks of lymph node metastasis. This personalized approach aims to identify regions where elective irradiation may be safely omitted, minimizing side effects without compromising tumor control. Material/Methods: We analyzed per-level lymph node involvement data from 392 OCSCC patients treated at two institutions: Centre Léon Bérard (France) and Inselspital Bern (Switzerland) [3]. All patients underwent neck dissection and lymph node involvement was determined pathologically. We trained a hidden Markov model (HMM) with this data to learn 17 highly interpretable parameters that represent lymphatic spread probability rates between tumor and lymph node levels (LNLs) and among LNLs. This model calculates the risk of occult disease in each LNL using clinical (imaging based) lymph node involvement, T-category, and mid-sagittal line extension. Based on a 5% risk threshold, these quantitative risk assessments where then used to investigate indications for elective irradiation in selected LNLs.

Results: The model achieved an accurate and precise fit to the observed data distributions, as visible in figure 1.