Ultrasound image texture analysis for liver fibrosis stage diagnostics

A comprehensive method of B-mode ultrasound image texture analysis for the determination of the liver fibrosis stage is suggested. The algorithm is based on the use of Rotation Forest and KNN classifiers for the texture classification. 720 textural characteristics were extracted using methods based on Laws’ masks analysis, co-occurrence matrix, gray level run-length matrix and statistical characteristics of the images. An optimal subset of 22 informative features was selected using correlation-based method. Testing the algorithm with liver images of 57 patients divided into 5 stages of fibrosis showed 72.7% classification accuracy for single regions of interest. In the case of entire image classification the fibrosis stage was correctly identified for the vast majority of cases.