Loading...
Researchers in Japan report findings of a convolutional three-dimensional neural network program trained to detect colon polyps based on short colonoscopy video segments. These were labeled frame by frame as containing polyp or no polyp by two experts. The test samples comprised 50 videos with polyp and 85 without polyp. The display indicated the presence of polyps by changing the color of the four corners of the endoscopic image to red when polyp was present.
In a frame-based analysis of the test samples, the sensitivity for polyp was 90%, specificity was 63%, and accuracy was 77%. In a polyp-based analysis, the sensitivity was 94% and false-positive detection was 60%.