Ulcerative colitis (UC) management relies on endoscopic scoring systems like the Mayo Endoscopic Subscore (MES) and UCEIS-but these are subjective, inconsistent, and often overlook disease heterogeneity. Inter-observer variability complicates both clinical decisions and trial outcomes.
Artificial intelligence is transforming this landscape. New video-based deep learning models analyze entire colonoscopy recordings, not just static images, enabling segmental, continuous, and reproducible disease assessment. Systems like Arges (Janssen R&D) and models from Byrne et al. demonstrate strong agreement with expert readers-kappa values exceeding 0.85 in multiple studies.
In the TITRATE trial re-analysis, AI uncovered significant treatment benefits missed by human reviewers, suggesting enhanced sensitivity for detecting therapeutic response. AI-driven metrics like the Cumulative Disease Score require fewer patients to demonstrate efficacy, streamlining clinical trials.
Challenges remain: real-world video quality, bowel prep variability, regulatory approval, and integration into clinical workflows. Most models are trained for activity scoring-not differential diagnosis-limiting use in atypical cases.
The future lies in hybrid human-AI systems. AI won’t replace gastroenterologists but will standardize scoring, support treat-to-target strategies, and bring objective data to community practices and global trials alike.