Development of a Novel Ulcerative Colitis Endoscopic Mayo Score Prediction Model Using Machine Learning.

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Endoscopic assessment is crucial for inflammatory bowel disease trials but suffers from observer variability.
• Standardizing endoscopic evaluation is essential for reliable clinical trial outcomes.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting endoscopic Mayo Score (eMS) in ulcerative colitis.
• To assess the ML model's ability to differentiate between inactive and active endoscopic disease using video data.

Main Methods:

• Utilized 793 full-length endoscopic videos from 249 ulcerative colitis patients in a clinical trial.
• Developed a video annotation approach to extract mucosal features and eMS labels for ML model training.
• Evaluated model performance on two independent test sets against human expert reads.

Main Results:

• The ML model achieved an AUC of 89% and 84% accuracy in differentiating inactive vs. active disease on the full test set.
• On the consensus test set, the model demonstrated higher performance with an AUC of 92% and 89% accuracy.
• The model showed strong predictive values, with positive predictive values of 80% and 87%, and negative predictive values of 85% and 90%.

Conclusions:

• A novel ML model accurately differentiates key levels of endoscopic disease activity in ulcerative colitis.
• The model, trained on video annotations and mucosal features, shows high performance in real-world clinical trial data.
• This approach offers a potential solution to reduce inter- and intraobserver variability in endoscopic assessments.