Automatic real-time capture and segmentation of endoscopy video

Endoscopy is a medical technology used to inspect the inner surface of organs such as the colon. During endoscopic inspection of the colon or colonoscopy, a tiny video camera generates a video signal, which is displayed on a monitor for manual interpretation by physicians. In practice, these images are not typically captured, which may be attributed by lack of tools for automatic capturing, automatic analysis of important contents, and quick and easy access to these contents. However, this lack of tools is being addressed by recent research efforts. This paper presents the description and evaluation results of novel software that automates the capture of all images of a single colonoscopy into a single digitized video file. The system uses metrics based on color and motion over time to determine whether the images are derived from inside a single patient. During testing our system extracted 173 videos totaling 70 hours of endoscopic video, out of 230 hours of raw video, with a segment-based sensitivity of 100% and specificity of 99%. No procedures were missed. Two video files contained only a non-patient video signal. The features of our system are robust enough to be suitable for day-to-day use in medical practice.