A computer vision model trained on 4,487 endoscopic images was able to identify mucosal ulcerations in Crohn’s disease patients with greater consistency than gastroenterologists and with strong correlation to the Simple Endoscopic Score for Crohn’s Disease (SES-CD). Published in Clinical Gastroenterology and Hepatology, the study highlights how AI can bring objectivity and reproducibility to colonoscopy interpretation — areas where manual scoring has long been inconsistent.
Key Takeaways
- Performance: AI achieved a higher similarity to ground truth (DICE score .591) than agreement between two gastroenterologists (.462).
- Correlation: Strong alignment with SES-CD, the current standard for scoring ulcer severity, but with more quantitative precision.
- Clinical Value: AI metrics could justify costly or high-risk therapies with more objective evidence.
- Access: In regions without IBD specialists, AI could provide expert-level guidance for treatment adjustments.
- Broader Impact: Potential applications in physician training, reducing inter-observer variability, and clinical trial standardization.
- Future Outlook: Seen as an early but crucial step toward automated Crohn’s disease care, with AI complementing human expertise.