Real-Time Video Analytics for Measuring Colonoscopy Performance

Abstract

We have developed a real-time video analytics system for measuring colonoscopy performance, including total visible surface areas, real-time feedback to the endoscopist about the measurements of visible surface areas, video clarity, preparation, and distention conditions. Our lab experiments show that the correlations between the computer and the experienced colonoscopist are: 76% in visible surface area estimation, 83.9% in clarity evaluation, 90% in preparation condition assessment, and 67.9% in distention condition evaluation. Our algorithm appears to be faster in response to dynamic scope movements compared to a 3D scope positioning device. In addition, the clinical experiment shows the system detected unexpected scope malfunction events in real-time.