Fully automated endoscopic disease activity assessment in ulcerative colitis

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Endoscopic assessment of ulcerative colitis (UC) is crucial but subjective.
• Subjectivity in endoscopic scoring can impact disease assessment accuracy and precision.

Purpose of the Study:

• To pilot a fully automated video analysis system for grading endoscopic disease in UC.
• To evaluate the potential of artificial intelligence (AI) in standardizing UC endoscopic scoring.

Main Methods:

• Developed AI models using convolutional neural networks to predict image quality and disease severity from UC endoscopic videos.
• Trained models on a developmental set of high-resolution videos with expert-assigned Mayo endoscopic scores (MESs).
• Validated the automated system on videos from a multicenter UC clinical trial, comparing AI scoring to human reviewer consensus.

Main Results:

• The AI system achieved high performance in classifying informative still images (sensitivity 0.902, specificity 0.870).
• Automated grading correctly predicted MESs in 78% of high-resolution videos and distinguished remission from active disease in 83.7% of clinical trial videos.
• AI scoring agreement with central reviewers improved to 69.5% when accounting for inter-reviewer variability.

Conclusions:

• Early results suggest AI can provide endoscopic disease grading in UC that approximates expert reviewer performance.
• The automated system holds potential for more objective and precise assessment of UC endoscopic disease.
• Further development could lead to AI-assisted endoscopy improving clinical trial efficiency and patient care.