Automated CRYPT Architecture Categorization Using Advanced Endoscopy

Le résumé fourni par la source

Alteration of mucosal structure of the colon is one of the major hallmarks of disease activity in inflammatory bowel diseases (IBD) such as Ulcerative Colitis (UC) and Crohn Diseases (CD). Ultra-magnification Endocytoscopy (ECS) enables real-time, in vivo assessment of mucosal microstructure, but its clinical application remains limited due to the complexity of image interpretation. This work focuses on the assessment of the crypt architecture of the colon mucosa using Endocytoscopy data leveraging recent foundational models in endoscopy, we classify ECS frames as showing normal crypt architecture or necrotic mucosa, domain aware pretrained ResNet-50 model with GastroNet weights is used to extract the features; using 5-fold cross-validation, we achieve a frame level classification accuracy of 79 % and when aggregating the frame scores to obtain a video-level classification, the overall accuracy obtained is 89 % with 83 % sensitivity, 90 % specificity and 91% AUC.

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