A deep learning algorithm distinguishing different endoscopic image content was trained to produce a predicted withdrawal time and extract relevant images. Comparison with video-based measurement of withdrawal time showed a median absolute difference of 0.4 minutes for AI prediction vs. 2.0 minutes for the standard reported withdrawal time. AI-generated documentation captured the cecum in 98/100 examinations vs. 88/100 for examiners’ original photodocumentation. Real-time capability was also demonstrated, suggesting its future potential for standardized reporting with reduced workload.
Quality: mobile360480720
Download
Publication History
Comments (0)