Development of an Automated Tool for the Estimation of Histological Remission in Ulcerative Colitis Using...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Ulcerative colitis (UC) management aims for histological and endoscopic remission.
• Current white light endoscopy (WLE) scores have limited correlation with histological activity.
• Single-wavelength endoscopy (SWE) visualizes microvascular changes, potentially improving assessment of histological remission.

Purpose of the Study:

• To evaluate the accuracy of a computer-aided diagnosis (CAD) system for estimating histological activity in UC.
• To compare the diagnostic performance of CAD systems based on WLE versus SWE.

Main Methods:

• A dataset of 6926 WLE and SWE frames from 112 UC patients was collected.
• Deep learning models (WLE-CAD and SWE-CAD) were trained to detect histological remission (Geboes score ≤ 2B.0).
• Model performance was assessed using sensitivity, specificity, and diagnostic accuracy.

Main Results:

• Initially, SWE-CAD (83.3% accuracy) outperformed WLE-CAD (67.5% accuracy) in detecting histological remission.
• Further training on the full dataset improved SWE-CAD performance to 95.2% accuracy.
• The final SWE-CAD model achieved 96.4% sensitivity and 92.9% specificity.

Conclusions:

• Automated CAD using SWE enhances capillary visibility, improving histological remission detection in UC.
• SWE-CAD achieved 95.2% diagnostic accuracy, offering objective assessment.
• This technology can help reduce inter-reader variability in UC assessment.