Application of Deep Learning Models to Improve Ulcerative Colitis Endoscopic Disease Activity Scoring Under Multiple...

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Endoscopic assessment of ulcerative colitis [UC] severity is limited by lack of clinical validation and inter-observer variability.
• Current scoring systems like the Mayo Endoscopic Subscore [MES] and Ulcerative Colitis Endoscopic Index of Severity [UCEIS] require expert interpretation.
• Automated tools are needed to improve the accuracy and efficiency of UC endoscopic scoring.

Purpose of the Study:

• To develop and validate a deep learning [DL] model for automated detection and severity scoring of ulcerative colitis [UC].
• To predict the Mayo Endoscopic Subscore [MES] and Ulcerative Colitis Endoscopic Index of Severity [UCEIS] using artificial intelligence.
• To accelerate the review process and improve the quality assurance of endoscopic assessments in UC.

Main Methods:

• A deep learning [DL] model was trained using 134 prospective endoscopic videos (1550030 frames) labeled by experts for MES and UCEIS.
• Convolutional neural networks [CNNs] with proprietary algorithms were employed for frame filtering, detection, and assessment.
• A graphical user interface was developed for video annotation and AI-driven disease severity display.

Main Results:

• The DL model demonstrated excellent accuracy in predicting MES and UCEIS, with low mean absolute error and bias.
• Strong agreement was observed between the AI model's predictions and expert labels, with quadratic weighted kappa values ranging from 0.78 to 0.90.
• The model achieved high performance at frame, section, and video levels for both MES and UCEIS scoring.

Conclusions:

• The developed tool is the first fully automated system for improving MES and UCEIS accuracy in ulcerative colitis [UC] assessment.
• The AI model significantly reduces the time required for video review and enhances subsequent quality assurance.
• This automated approach offers a promising solution to overcome limitations in current endoscopic scoring of UC.