Fully automatic segmentation and classification of liver ultrasound images using completed LBP texture features

Abstract

In this paper, a novel method is presented to discriminate fatty, normal and heterogeneous livers based on textural analysis of liver ultrasound images using Completed Local Binary Pattern (CLBP). The proposed approach is able to fully automatically select the optimum regions of interest (ROIs) of the liver images. These optimum ROIs are analyzed to extract CLBP features. A support vector machine (SVM) classifier is then employed to classify the fatty, normal and heterogeneous livers. The fully automatic scheme to select the ROIs with low computational cost and CLBP to extract texture features clearly illustrates the efficiency of this system. The results showed the overall accuracy of 98.67% with sensitivity of 100% for fatty and heterogeneous class.