AI-based quantification of inflammatory extent for relapse prediction in ulcerative colitis: a prospective cohort...

Area of Science:

• Gastroenterology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Current ulcerative colitis (UC) assessments focus on inflammation severity, not spatial extent.
• Predicting clinical relapse in UC patients in remission remains challenging.
• Novel methods are needed to improve prognostic accuracy in UC management.

Purpose of the Study:

• To develop and validate a deep learning-based score (QUAD) for UC assessment.
• To evaluate if incorporating inflammatory extent improves UC relapse prediction.
• To assess the utility of AI-derived inflammatory extent in predicting clinical relapse in UC.

Main Methods:

• A prospective cohort study involving UC patients in clinical remission.
• Development of the Quantitative Ulcerative Colitis Assessment using Deep Learning (QUAD) score (0-12) using 84,743 images from 998 patients.
• Validation using still image and automated full-length video analysis for relapse prediction over 24 months.

Main Results:

• Patients with QUAD score ≥4 had a significantly higher relapse rate (19.4%) than those with QUAD <4 (5.2%).
• Automated video analysis of the distal 10% colon segment showed the highest predictive performance (AUC 0.73).
• AI-driven assessment of inflammatory extent demonstrated superior relapse prediction compared to whole-colon or still image analysis.

Conclusions:

• AI-augmented assessment integrating inflammatory severity and extent offers valuable prognostic information in UC.
• The QUAD score and AI-based video analysis may enhance prediction of clinical relapse in UC patients.
• Inflammatory extent, particularly in the distal colon, is a critical factor in UC relapse prediction.