Automated AI-Based Colonoscopy Assessment for Ulcerative Colitis

K Hammouda et al. Journal of Crohn’s and Colitis, Volume 20, Issue 8, August 2026, jjag122. Open Access! Automated AI-based Mayo Endoscopic Scoring for ulcerative colitis across adult and pediatric cohorts from diverse populations

Background: “Endoscopic evaluation remains the gold standard for assessing UC disease activity…The Mayo Endoscopic Score (MES), ranging from 0 to 3, is a widely used index to grade mucosal disease severity in trials and clinical practice, with scores of 0-1 indicating remission and 2-3 reflecting active disease.6 However, the clinical utility of MES is limited by substantial inter- and intra-observer variability7…By leveraging large, annotated datasets, DL models can automate MES grading, potentially reducing subjectivity and enhancing diagnostic consistency.11“

Methods: “Using three multi-institutional datasets, we trained and validated convolutional neural network models to predict MES. The primary dataset (LIMUC; 564 adults, 11 276 images, Turkey) supported model development, with external validation on TMC (308 adults, 7978 images, China) and the Emory Pediatric dataset (EPC; 80 children, 113 images, USA).”

Key findings:

  • UC-Re for binary remission classification (MES 0-1 vs 2-3): “Achieved AUCs of 0.98, 0.95, and 0.98 across LIMUC, TMC, and EPC”
  • UC-MES for four-class grading (MES 0-3): “Demonstrated strong ordinal consistency (QWK = 0.81-0.85), comparable to inter-expert agreement (QWK = 0.88)”
(A) ROC curves for the binary UC-Re task (active vs inactive disease) in the
Emory, LIMUC, and TMC test sets (AUC 0.974, 0.980, and 0.949; average 0.97)
D) UC-MES four-class model (MES 0-3) compared with two board-certified gastroenterologists on the Emory cohort, measured by accuracy, F1-score, and quadratic weighted κ (QWK). UC-MES, Mayo Endoscopic Subscore grading; QWK, quadratic weighted kappa.

My take: This study while narrowly focusing on assessment of UC activity shows the potential of an AI-based framework to improve/standardize the interpretation of colonoscopy/endoscopy interpretation. This could help as a decision support tool and could facilitate comparison to prior evaluations. Presumably with adequate training and proof with prospective trials, these findings could be extended to a broad range of GI disorders that utilize endoscopic monitoring.

Related blog posts:

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.