Automatic Quantification of Crypt Architecture in Ex Vivo Gastrointestinal Epithelium for High-Resolution Microendoscopic Images

摘要

High-resolution microendoscopy based on the fiber bundle has been showing immense potential to early detection of precancerous and cancerous lesions in gastrointestinal epithelium, especially for low-resource areas in China. However, obtaining clinical benefit from microendoscopic diagnosis usually remains in the hands of experts. Quantitative analysis focusing on computer-aided detection is therefore receiving attention as an attractive tool. In this paper, we present an automatic quantification method of crypts in gastrointestinal epithelium for high-resolution microendoscopic images, which is composed of four modules: filtering, contrast enhancement, crypt segmentation and morphologic quantification of crypts. The preliminary experiments on ex vivo image data indicate that the proposed method is effective for crypt segmentation from microendoscopic images with low-contrast, and quantitation of well-defined clinical features, which has a potential in future computer-aided diagnostic systems by revealing the morphologic characteristics of crypts at various clinical stages. The proposed method also enables instant processing. Thus, it may be a powerful tool for assisting endoscopists in real-time interpretation of high-resolution microendoscopic images, with high accuracy and consistent diagnosis. Furthermore, we are testing the method on larger gastrointestinal epithelium images and in vivo high-resolution microendoscopic images, and will integrate this work into a computer-aided diagnostic system.