Influence of ultrasound despeckling on the liver fibrosis classification

Abstract

An analysis of speckle filtering influence on B-mode ultrasound image texture-based determination of the liver fibrosis stage has been performed. We developed a comprehensive method for liver texture analysis based on 10–20 textural characteristics. These characteristics were found as most informative from 1390 textural features calculated using Laws' masks, co-occurrence matrix, gray level run-length matrix, wavelets and statistical characteristics of the images. We used Siemens ACUSON S2000 ultrasound images of liver cuts along the right midclavicular line for more than 50 patients for fibrosis classification using the METAVIR score. The classification was performed using Multi-layer Perceptron, Random Forests and KNN classifiers with data balancing using SMOTE algorithm. The ultrasound despeckling was performed using SRAD algorithm with an entropy-based stopping criterion. It was found that speckle filtering procedure enhances the classification and increases AUROC value by 5%.