ESTRO 2025 - Abstract Book

S3986

Radiobiology - Tumour radiobiology

ESTRO 2025

Results: Our model achieved DSC of 0.82±0.09 and 0.71±0.10 on the test set and additional test set, respectively, outperforming the baseline nnUNetV2, which had DSC of 0.79±0.10 and 0.66±0.11. Similarly, ASSD of the MedNext model was superior to the nnUNetV2 in both the test set (0.92±0.56 versus 1.06±0.64) and the additional test set (1.30±0.26 versus 1.64±0.55). Figure 2 illustrates the decreasing trend of predicted HMRs on periodic FBCT, indicating progressive tumour shrinkage during radiotherapy.

Conclusion: By employing the seven-layer 3D MedNext model, HMRs can be delineated accurately on CT images both before and during radiotherapy. This work represents the first attempt to visualize high metabolic regions of NPC without PET as input. Such an approach may guide the optimization of irradiation fields and dosing in a CT-only setting.

Keywords: nasopharyngeal carcinoma, high metabolic regions

References: 1. Meng M, et al. DeepMTS: Deep Multi-Task Learning for Survival Prediction in Patients With Advanced Nasopharyngeal Carcinoma Using Pretreatment PET/CT. IEEE J Biomed Health Inform. 2022;26(9):4497-4507. doi:10.1109/JBHI.2022.3181791 2. Roy S, et al. MedNeXt: Transformer-driven Scaling of ConvNets for Medical Image Segmentation. 2023:arXiv:2303.09975. doi:10.48550/arXiv.2303.09975 Accessed March 01, 2023. https://ui.adsabs.harvard.edu/abs/2023arXiv230309975R 3. De Bruycker A, et al. Disease Control and Late Toxicity in Adaptive Dose Painting by Numbers Versus Nonadaptive Radiation Therapy for Head and Neck Cancer: A Randomized Controlled Phase 2 Trial. Int J Radiat Oncol Biol Phys. 2024;120(2):516-527. doi:10.1016/j.ijrobp.2024.01.012