Efficacy of an AI System in Training Endoscopists to Assess Gastric Intestinal Metaplasia Via the EGGIM Score

This prospective randomized controlled trial with a crossover design incorporated image-enhanced endoscopy (IEE) videos demonstrating complete standardized examinations of five standard gastric areas (antrum greater curvature, antrum lesser curvature, incisura, corpus lesser curvature, and corpus greater curvature). Endoscopists were stratified by experience level and randomly assigned to either the AI-assisted scoring first group, which performed EGGIM scoring with AI assistance in the initial phase followed by conventional scoring after a washout period, or the conventional scoring first group, which completed the assessments in reverse order. The study primarily evaluated the training efficacy of the EGGIM-AI system for improving endoscopists' EGGIM scoring performance by comparing diagnostic accuracy metrics, including the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity between groups at different study phases.

Inclusion Criteria: Endoscopists who have performed ≥50 image-enhanced endoscopy (IEE) procedures per year and demonstrated competency in performing standardized IEE. Exclusion Criteria: Endoscopists who participated in data acquisition or were unblinded to patients' identifiable information and clinical data.