Hepatic Steatosis detection using the co-occurrence matrix in tomography and ultrasound images

Abstract

Hepatic Steatosis (HS) or Fatty Liver is a disease due to fat accumulation within hepatocytes. This disease requires treatment to avoid clinical complications such as hepatic inflammation, fibrosis and finally chronic hepatic damage and hepatic carcinoma. An algorithm for performing the manual segmentation was used. A polygon is traced for representing the region of interest in tomography (CT) images as well as in Ultrasound (US) images. These regions are then subdivided in a set of windows of size 4×4. For each of the windows the co-occurrence matrix is estimated as well as several descriptive statistical parameters. From these matrices, 9 descriptive statistical parameters were estimated. A Binary Logistic Regression (BLR) model was fitted considering as dependent variable the presence or absence of the disease and the descriptive statistical parameters as predictor variables. The model attains classification results of HS with a sensibility of 95.45% in US images and 93.75% in CT images in the venous phase.