Extracting clinical information from endoscopic capsule exams using MPEG-7 visual descriptors

The endoscopic capsule is a recent technological breakthrough with high clinical importance. Exam analysis duration is its main setback, requiring an average of two hours from a trained specialist. Automation is required and this paper presents a topographic segmentation tool using low-level features that can reduce annotation times up to 15 minutes per exam. This is accomplished using Bayesian classifiers and MPEG-7 visual descriptors.