High Accuracy in Classifying Endoscopic Severity in Ulcerative Colitis Using Convolutional Neural Network.

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Endoscopic assessment is vital for ulcerative colitis (UC) management.
• Current endoscopic evaluations have significant variability, impacting reliability.
• Standardized, objective assessment of UC endoscopic severity is needed.

Purpose of the Study:

• To develop a deep learning model for automated UC endoscopic severity evaluation.
• To differentiate between active and healed mucosa in UC patients.
• To classify different degrees of endoscopic disease severity using the Mayo endoscopic subscore (MES).

Main Methods:

• Utilized 1,484 endoscopic images from 467 UC patients.
• Images were independently classified by two experts using the MES, with a third expert resolving disagreements.
• Convolutional neural networks were trained and validated using five-fold cross-validation.

Main Results:

• The model achieved 0.84 accuracy in distinguishing all four MES categories.
• Binary classification tasks showed high accuracy: 0.94 for MES 0 vs 1-3 and 0.93 for MES 0-1 vs 2-3.
• Areas under the ROC curve were excellent, reaching 0.998 for binary tasks.

Conclusions:

• A novel, highly accurate deep learning model for automated UC endoscopic image evaluation has been developed.
• The model effectively distinguishes all four MES levels of disease activity.
• This automated approach promises to optimize and standardize MES evaluation, improving consistency across healthcare settings.