ESTRO 2025 - Abstract Book

S2473

Physics - Autosegmentation

ESTRO 2025

Results: Supervised segmentation results, presented in Table 1, were obtained by ensembling predicted masks from nnU-Net and MedNeXt [5] models, selected for their superior performance observed in our experiments.

The proposed rule-based pipeline achieved a Dice score of 0.82 on the clinical dataset, demonstrating its potential for effective segmentation. Comparatively, the supervised model yielded a Dice score of 0.81, indicating a marginal (1%) decrease relative to the pseudo-references. This suggests comparable performance between the supervised and rule-based approaches in extending GTV boundaries while preserving sensitive tissues, strengthened by the high agreement (Dice=0.95) between predicted CTV masks and pseudo-references on the BraTS dataset. Conclusion: These promising preliminary results highlight the potential of automated techniques for segmenting CTV in glioma. Future studies will incorporate large-scale clinical labelled datasets for evaluation of the supervised approach. References: [1] M. Niyazi et al. , ESTRO-EANO guideline on target delineation and radiotherapy details for glioblastoma, Radiotherapy and Oncology , 10.1016/j.radonc.2023.109663. [2] M. C. de Verdier et al. , The 2024 Brain Tumor Segmentation (BraTS) Challenge: Glioma Segmentation on Post treatment MRI, https://arxiv.org/abs/2405.18368v1 [3] B. Billot et al. , “SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining,” Med Image Anal , 10.1016/J.MEDIA.2023.102789. [4] S. Zolotova, et al., Burdenko-GBM-Progression | Burdenko’s Glioblastoma Progression Dataset, 10.7937/E1QP D183 Keywords: Clinical target volume, segmentation, glioma