Intelligent Diagnosis of Non-alcoholic Fatty Liver Disease Based on Multi-modal Ultrasound Features

Non-alcoholic fatty liver disease (NAFLD) is a popular chronic disease worldwide, with a wide spectrum of conditions, ranging from fatty liver (NAFL) to non-alcoholic steatohepatitis (NASH). Liver biopsy is the gold standard to diagnose NAFLD. However, it is an invasive procedure that is prone to sampling error and can lead to complications, which makes it an urgent need to develop non-invasive tests. Ultrasound (US) is an inexpensive, nonionizing-radiative, and real-time imaging modality that is used widely in practice. Ballooning and inflammation are very important for evaluating NASH resolution. In this study, we stage ballooning and inflammation based on multi-modal US features extracted from radiofrequency (RF) data, B-mode images, shear wave elastography (SWE) images, and baseline data. The results demonstrated that the multi-modal quantitative US features can stage ballooning and inflammation with moderate or good performance, indicating that the multi-modal US features may be a promising way to guide liver biopsy and NASH resolution.