Prior‐guided automatic delineation of post‐radiotherapy gross tumor volume for esophageal cancer

Le résumé fourni par la source

BACKGROUND: Integrating post-radiotherapy (RT) CT into longitudinal esophageal cancer response models substantially improves predictive accuracy. However, manual delineation of gross tumor volume (GTV) on post-RT CT is both labor-intensive and time-consuming. PURPOSE: We propose a novel deep learning-based framework that integrates medical physics priors-pre-RT GTV contours and radiotherapy dose distributions-to automatically delineate post-RT GTV. METHODS: A multicenter retrospective cohort of 294 EC patients (225 training, 45 internal validation, 24 external validation) was assembled. Pre-RT CT scans, GTV contours, and dose map were co-registered and cropped to 256 × 256. We implemented an nnU-Net v2 backbone, incorporating high dose region and pre-RT GTV priors via element-wise multiplication and element-wise addition to guide feature extraction. Performance was evaluated using anatomical (Dice, IoU, HD95, ASSD, Precision, Recall) and radiomics analyses (ICC, Pearson correlation, LASSO-Cox, C-index) across internal and external cohorts. RESULTS: In cross-validation, the optimal fold achieved DSC = 0.7809 ± 0.1310, IoU = 0.6486 ± 0.1507, HD95 = 3.6321 ± 2.0942, and ASSD = 1.9673 ± 1.0352 (p < 0.0167 vs. state-of-the-art models). Ablation studies demonstrated that combining two types of medical physics priors outperformed single-prior or no-prior models (internal: DSC = 0.7723 ± 0.1290; external: DSC = 0.7545 ± 0.1058). Radiomic features extracted from automated contours exhibited high reproducibility (78.6% with ICC > 0.75) and strong concordance with manual features (R > 0.8), yielding comparable prognostic performance (C-index Δ nonsignificant). CONCLUSION: By embedding medical physics priors into a self-configuring nnU-Net v2, our method achieves accurate and robust automated delineation of post- RT GTV in EC across multiple centers. This approach has potential to facilitate the construction of treatment response prediction models.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.