Barrier healing is emerging as a key therapeutic target in Inflammatory bowel disease (IBD), being more closely associated with sustained remission and improved long-term outcomes than endoscopic and histological remission alone. Accurate assessment of barrier integrity is therefore crucial to optimise disease management. Probe-based confocal laser endomicroscopy (pCLE) enables real-time structural and functional assessment of intestinal barrier. However, its interpretation remains complex and highly operator-dependent, limiting its reproducibility and widespread application. Preliminary studies have demonstrated the feasibility of using automated assessment of specific pCLE features to predict therapeutic response¹. We aimed to develop and validate an artificial intelligence (AI)-driven pCLE system for standardised, objective assessment of barrier impairment (BI) in IBD.

156 high-quality pCLE videos from IBD patients undergoing colonoscopy in Ireland and UK were considered. Experienced endoscopists assessed the presence of BI according to a previously developed pCLE-scoring system², providing a reference standard for AI. An AI algorithm was developed to predict pCLE-based BI using a weakly supervised learning paradigm that exclusively employs video-level scoring. Frame-level features were extracted with ViT pretrained under DINOv2 self-supervised learning framework³ and transformer-based architecture⁴ aggregated these into a global video representation. A linear layer classifier with two output neurons generated the final video-level BI score. A total of 105 videos were used for training and validation of the framework, while 51 were reserved for testing. The model’s ability to predict overall, epithelial, and vascular BI was evaluated. Diagnostic performance was expressed as sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and F1 score.

Overall, 48 IBD patients (28 CD and 20 UC) were considered. 119/156 (76%) videos showed BI, 68/101 (67%) in Ireland and 51/55 (93%) in the UK cohort. Table 1 details the diagnostic performance of the model. The AI algorithm achieved 77% sensitivity, 75% specificity, and 77% accuracy in detecting overall BI at the video level. Moreover, the model assessed the presence of epithelial and vascular BI with 70% and 85% sensitivity, 72% and 77% specificity, 71% and 82% accuracy. Diagnostic performance was higher in UC than CD.

Our AI model enables automated and standardised assessment of intestinal impairment in IBD, offering an objective and reproducible tool to monitor barrier healing as a novel therapeutic endpoint.