Improving the Computer-Aided Estimation of Ulcerative Colitis Severity According to Mayo Endoscopic Score by Using...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Endoscopic assessment of ulcerative colitis (UC) activity is crucial for patient management but suffers from significant inter- and intraobserver variability.
• This variability can impact treatment decisions and disease monitoring.
• A reliable and objective method for assessing endoscopic activity is needed.

Purpose of the Study:

• To develop a computer-aided diagnosis (CAD) system using deep learning to objectively assess endoscopic activity in UC.
• To reduce subjectivity and enhance the reliability of endoscopic evaluations.
• To improve the accuracy of grading disease severity according to the Mayo endoscopic score (MES).

Main Methods:

• A cohort of 11,276 colonoscopy images from 564 UC patients was utilized.
• A regression-based deep learning approach was proposed for endoscopic evaluation.
• Five state-of-the-art convolutional neural network (CNN) architectures were compared using ten-fold cross-validation.

Main Results:

• Classification-based CNNs showed excellent agreement with expert annotations for UC endoscopic scoring.
• The proposed regression-based deep learning approach significantly improved the performance of most models.
• The regression-based models demonstrated increased robustness across different cross-validation folds.

Conclusions:

• Deep learning, particularly CNNs, demonstrates strong performance in evaluating UC endoscopic activity.
• Integrating domain knowledge into CNNs further enhances performance and robustness.
• This approach shows promise for accelerating the clinical translation of AI in UC management.