Diagnosing of fatty and heterogeneous liver diseases from ultrasound images using fully automated segmentation and hierarchical classification

In this paper, a fully automatic approach to select the regions of interest (ROIs) of the liver images and an automatic hierarchical procedure to characterize normal, fatty and heterogeneous livers, using textural analysis of liver ultrasound images are described. The proposed algorithm contains two stages. The first stage, automatically assigns some ROIs in a liver ultrasound. In the second stage, discrimination between heterogeneous, fatty and normal livers is performed in a hierarchical method. This stage, first, classifies focal and diffused livers and then discriminates fatty and normal ones. The wavelet packet transform is used to analyze liver texture and obtaining a number of statistical features. A support vector machine classifier is employed to classify three classes. The fully automatic scheme to select the ROIs with low computational cost and the hierarchical classification scheme outperformed the non-hierarchical one-against-all schemes, achieving an overall accuracy of 97.9%.