Identification of advanced hepatic steatosis and fibrosis using ML algorithms on high-frequency ultrasound data in patients with non-alcoholic fatty liver disease

Abstract

Liver diseases are an ever-growing global problem. Liver fibrosis or liver steatosis are often observed accompanying liver diseases. Currently, transient elastography is often used as a non-invasive tool to assess liver health but the corresponding equipment is comparatively complex and expensive. In this work, we provide preliminary results showing how one-dimensional ultrasound radio-frequency signals can be used for the non-invasive diagnosis of liver fibrosis and liver steatosis by deploying various Machine Learning algorithms. We show that a SVM model performing on Wavelet transformed ultrasound radio-frequency signals yields the best performance for fibrosis stage assessments (with an average F1 score of 85.71 %) and steatosis stage assessments (with an average average F1 score of 80.95 %).