earticle

영어

BilIN (Biliary INtraepithelial Neoplasm) is a precursor lesion of Intrahepatic CholangioCarcinoma (ICC). It is important to diagnose ICC and distinguish its stages for the prevention of ICC and proper treatment for a patient. BilIN is classified into BilIN-1, BilIN-2, and BilIN-3 by the morphological change and loss of polarity of epithelial cell and the structural abnormality of epithelium. This paper proposes two quantitative features based pathological knowledge for distinguishing stages of BilIN. The first feature is LumenBoundaryAbnomality (LBA) to measure the abnormal structure of epithelium; the second feature is NucleiPolarity for determine loss of polarity of epithelial cells. The experiment performed the stage classification of BilIN using following features; non-epithelial nuclei features, epithelial nuclei features, proposed features, non-epithelial nuclei and proposed features, epithelial nuclei and proposed features. The classification learning algorithm is used back-propagation Artificial Neural Networks. The classification result showed classification accuracies of 35% with non-epithelial features, 40% with epithelial features, 74% with proposed features, and 35% with non-epithelial and proposed features, and 46% with epithelial and proposed features.