Using a Deep Learning Model to Address Interobserver Variability in the Evaluation of Ulcerative Colitis (UC)...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate assessment of ulcerative colitis (UC) severity via endoscopic images is vital for treatment decisions.
• Interexpert variability in interpreting endoscopic images leads to inconsistent UC diagnoses.
• Existing methods lack standardization, impacting clinical reliability.

Purpose of the Study:

• To develop and validate a deep learning-based standardization method for UC endoscopic image assessment.
• To address interobserver variability in classifying UC severity using the UCEIS scoring system.
• To establish a reliable consensus model for objective UC severity evaluation.

Main Methods:

• Collected 254 rectal endoscopic images from 115 UC patients.
• Five experts assigned UCEIS scores, followed by interobserver variance analysis.
• Developed a consensus model based on majority expert agreement.
• Validated the model on 50 independent endoscopic images.

Main Results:

• Interobserver variance showed an ICC of 0.8431 for UCEIS scores.
• The consensus model achieved 0.94 accuracy on test images.
• Consensus model outperformed models trained on individual expert labels.

Conclusions:

• A deep learning consensus model effectively standardizes UC endoscopic image interpretation.
• This approach significantly reduces diagnostic variability and enhances assessment accuracy.
• The developed model offers a reliable tool for objective UC severity determination.