ESTRO 2025 - Abstract Book

S1027

Clinical – Head & neck

ESTRO 2025

region as described above. Model performance was compared to a clinical benchmark model: HNC-PREDICTOR[2]. Hybrid models used both predictions from HNC-PREDICTOR and image model as predictors in a Cox model.

Results:

The 2D ResNet18 models using T2 images within a bounding box region determined by GTV volume with a 5-mm margin (T2_GTV_cmargin) achieved notably higher C-index values of 0.88 and 0.75 for LC and OS prediction, respectively, compared to 0.68 and 0.70 achieved by CT-based models and 0.61 and 0.67 by PET models (Figure 2). In comparison to the clinical benchmark model, the T2-based model showed improved LC prediction (C-index: 0.88 vs. 0.80) and combining the T2 and clinical model (hybrid models) resulted in an improved OS prediction (C-index 0.81 vs. 0.78). Furthermore, the combined clinical/T2 models showed good risk stratification ability and a good calibration. Conclusion: CNN models including MRI information improved predictive performance compared to the clinical benchmark models and CT- or PET-based models for LC and OS. Models with the best performance were MRI-only models for LC and hybrid models including both MRI and clinical features for OS.

Keywords: MRI; Outcome Prediction; Clinical model

References: [1]

He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., vol. 2016- December, 2016. https://doi.org/10.1109/CVPR.2016.90. [2] van Dijk L V, Mohamed ASR, Ahmed S, Nipu N, Marai GE, Wahid K, et al. Head and neck cancer predictive risk estimator to determine control and therapeutic outcomes of radiotherapy (HNC-PREDICTOR): development, international multi-institutional validation, and web implementation of clinic-ready model-based risk stratification. Eur J Cancer 2023;178:150–61.