Description and Classification of Confocal Endomicroscopic Images for the Automatic Diagnosis of Inflammatory Bowel Disease

摘要

Confocal Endomicroscopy (CEM) is a newly developed diagnosis tool which provides in vivo examination of the gastrointestinal (GI) histological architecture, avoiding the traditional biopsy. The analysis of CEM images is a challenging task for experts, since there isn't a clearly defined taxonomy of the several disease stages. We aim at building an automatic on-the-fly classifier to provide useful clinical advices for diagnosis. In this work, we propose to make a split between two main subsets of our expert-annotated database: low and high probability of pathology. We focus on segmentation techniques to extract relevant histological structures, and then encode this information in a feature vector used for classification.