Machine Learning Models for the Assessment of the Mayo Endoscopic Score in Ulcerative Colitis Trial Endpoints: A...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• The Mayo endoscopic score (MES) is crucial for ulcerative colitis (UC) clinical trials but suffers from subjective human assessment.
• Machine learning (ML) offers a potential solution to standardize MES evaluations and improve reproducibility.
• Understanding ML model performance is key for broader application in UC research.

Purpose of the Study:

• To systematically review the training and testing of ML models for predicting MES from endoscopic videos in UC patients.
• To evaluate the performance characteristics of ML-based MES prediction models.

Main Methods:

• Systematic literature search of PubMed/MEDLINE, EMBASE, and Web of Science up to December 31, 2024.
• Inclusion of studies training or testing ML models for automated MES grading on UC endoscopic videos.
• Reference checks and Google searches supplemented the primary database search.

Main Results:

• Seven studies met inclusion criteria; five reported model performance.
• Accuracy for predicting ordinal MES grades (0-3) ranged from 56.8% to 83.3%.
• Accuracy for predicting dichotomized MES (improvement/remission definitions) ranged from 84% to 95.5%.

Conclusions:

• ML models demonstrate robust performance for MES assessment in UC endoscopic videos.
• This technology presents a standardized and reproducible approach to measuring endoscopic severity.
• Further research is needed to assess the impact of ML-based MES assessment on clinical trial outcomes.