抄録
In histopathological microscopic images, the morphological structure of nuclei plays an important role
to diagnose a cancer . In addition, multispectral images may provide more significant information than RGB images for nucleus analysis. In this paper, we present the signific ance of multispectral data in terms of classifying cancer and non cancer cells. We propose a methodology to identify cancer cells of hepatocellular carcinoma (HCC) by
extracting nuclear textures from 100x multispectral images of HE stained tissues. The pro posed system is composed of 2 stages: nuclei segmentation and nuclei classification based on their textural features. Precisely, pixel based classification is performed using a random forests classifier to automate the nuclei segmentation.Subsequently, nu clei textural features are extracted using multifractal descriptors. We utilized a texton based method to extract the texture of the nuclei and classify normal and cancer cells. Experimental results show that our method can automatically classif y the cell nuclei.
