From diagnosis to prognosis: artificial intelligence–driven endoscopy for precision care in inflammatory bowel disease
Abstract
Artificial intelligence (AI) is reshaping the diagnostic and prognostic landscape of inflammatory bowel disease. The capacity of AI to standardize endoscopic activity scoring and reduce inter-observer variability is now well-established: agreement between convolutional neural networks and expert readers (kappa, 0.84) is comparable to that between two experts (kappa, 0.86), and models achieve an area under the curve of approximately 0.97 for distinguishing remission from active disease. However, the most transformative potential lies in prognostication. A central application in this regard is the non-invasive prediction of histologic remission, which addresses the endoscopy–histology discordance that currently limits risk stratification. Advanced models report sensitivities of 65%–98% and specificities of 80%–97% for this virtual histology task. Longitudinal data show that AI-based assessments of histologic or vascular activity can strongly predict clinical relapse, with twelve-month relapse rising from approximately 3% to 25% and hazard ratios reaching nearly nine. Fusion models that integrate endoscopic and digital pathology data achieve 93% accuracy for histological remission and outperform single-modality systems, laying the foundation for an endo-histo-omics approach to therapy selection. Despite the persistent challenges in validation, generalizability, and reporting standards, this diagnosis-to-prognosis-to-precision pathway represents the future of individualized management of inflammatory bowel disease.