Artificial Intelligence in Cholangioscopy: A New Frontier for Indeterminate Biliary Strictures

Indeterminate biliary strictures remain difficult to diagnose because benign and malignant conditions may have overlapping clinical, radiological, and cholangioscopic appearances. Conventional endoscopic retrograde cholangiopancreatography (ERCP)-based sampling has high specificity but limited sensitivity, often leading to repeated procedures or prolonged follow-up. Digital single-operator cholangioscopy (D-SOC) improves assessment by enabling direct visualization and targeted biopsy, but its accuracy remains limited by operator-dependent interpretation, variable image quality, interobserver variability, and sampling error. Artificial intelligence (AI), particularly deep learning-based image analysis, may enhance cholangioscopy by supporting real-time malignancy prediction, identifying suspicious mucosal and vascular features, guiding targeted biopsy, standardizing reporting, and assisting training. Early studies suggest that AI models can differentiate malignant from benign biliary strictures using digital cholangioscopy images and recognize features, such as tumor vessels, papillary projections, nodularity, ulceration, and mass-like mucosal change. However, AI-assisted cholangioscopy remains at an early stage. Current evidence is limited by retrospective designs, selected image datasets, small sample sizes, variable reference standards, and limited external validation. Prospective multicenter studies using real-time video are needed to determine whether AI improves diagnostic yield, reduces false-negative biopsies, and shortens time to diagnosis. At present, AI should be viewed as an adjunct to expert cholangioscopic assessment, histology, imaging, and multidisciplinary review rather than a replacement.