ESTRO 2025 - Abstract Book

S3786

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

the deep learning model by incorporating additional imaging data such as PET and investigating the insights provided by the attention maps.

Keywords: Radiation-pneumonitis, NTCP model, Deep learning

References: [1] Niezink, Anne GH, et al. "External validation of NTCP-models for radiation pneumonitis in lung cancer patients treated with chemoradiotherapy." Radiotherapy and Oncology 186 (2023): 109735

2480

Proffered Paper Towards a multi-toxicity NTCP modelling approach for head and neck cancer

Daniel Connor C MacRae 1 , Luuk van der Hoek 1 , Suzanne P.M. de Vette 1 , Hendrike Neh 1 , Peter M.A. van Ooijen 1 , Nanna M Sijtsema 1 , Amy C Moreno 2 , Clifton D Fuller 2 , Johannes A Langendijk 1 , Matias Valdenegro Toro 3 , Lisanne V van Dijk 1 1 Radiation Oncology, Universitair Medisch Centrum Groningen, Groningen, Netherlands. 2 Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, USA. 3 Bernoulli Institute, University of Groningen, Groningen, Netherlands Purpose/Objective: Modern normal tissue complication probability (NTCP) models for patients with head and neck cancer (HNC) are trained to predict the risk of a single toxicity (single-tox) at a time [1]. This approach fails to account for interactions between different toxicities, despite evidence suggesting relationships between the development of different toxicities (Figure 1) [2], [3]. Deep learning (DL) models have the potential to capture information about these toxicity interactions as they can be trained to predict multiple toxicities simultaneously. This study aims to improve predictive performance for each toxicity by enabling a multi-toxicity (multi-tox) DL NTCP model to leverage knowledge about these inter-toxicity relationships.