Hybrid patch-based and image-wide classification of confocal laser endomicroscopy images in Barrett's esophagus surveillance

Abstract

Barrett's esophagus (BE) is a premalignant condition characterized by the replacement of normal squamous esophageal epithelium by metaplastic intestinal epithelium containing goblet cells. To be diagnosed and monitored, BE requires a thorough observation of epithelial macro- and microscopic changes. Confocal laser endomicroscopy (CLE) has recently revealed to be a useful technique for in vivo virtual histology for BE surveillance. We present a computer-based method for the automatic classification of gastric metaplasia (GM), intestinal metaplasia (IM) and neoplasia (NPL) on the basis of appearance features of confocal images. Comparing the automatic results with the histological gold standard, the proposed method classifies IM, GM, and NPL confocal images with accuracy comparable to human observer. Moreover, it increases the sensitivity and the specificity of CLE examinations, thus decreasing the number of biopsies needed for BE and neoplasia diagnosis.