An Effective Automated Grading System for HCC in Biopsy Images

Accurate grading for hepatocellular carcinoma (HCC) in biopsy images is important to prognosis and treatment planning. However, visual grading is always time-consuming, subjective, and inconsistent. In this paper, we proposed a novel approach to automatically classifying biopsy images into five grades. At first, a dual morphological reconstruction method was applied to remove noise and accentuate nuclear shapes. Then we used watershed and snake techniques to smoothly segment nuclei from their background. Fourteen features were extracted according to six types of characteristics. We constructed a hierarchical classifier using Support Vector Machine and Sequential Floating Forward Selection method to automatically select an optimal set of features at each decision node of the classifier. Our experimental results demonstrated that 94.5% of accuracy can be achieved for a set of 604 biopsy images. Key-Words: Hepatocellular Carcinoma, Morphological Reconstruction, Watershed, Snake, Support Vector Machine, Sequential Floating Forward Selection, Hierarchical Classification