Intelligent Grading of Liver Fibrosis, Inflammation and Steatosis Using Handcrafted and Deep Features From Multimodal Ultrasound Data

Liver fibrosis, inflammation and steatosis are three typical pathological forms of chronic liver diseases (CLDs) and affect lots of people worldwide. The early diagnosis and accurate grading of these three diseases are decisive for controlling the development of the diseases, promoting the reversal of the disease course and improving the prognosis of patients with CLDs. This study proposes an intelligent grading framework for these three diseases based on a multi-parametric quantitative ultrasound (QUS) method to extract both explicable handcrafted and high-dimensional deep features from multimodal ultrasound data, including B-mode images, shear wave elasticity (SWE) images and radio-frequency and envelope signals of transient elastography (TE). The handcrafted and deep features are fused and used as inputs of an artificial neural network (ANN) to grade liver fibrosis, inflammation and steatosis simultaneously. With accuracy (ACC), sensitivity (SEN), specificity (SPE) and area under the receiver operating characteristic curve (AUC) as the evaluation indexes, the proposed method performs well with all the evaluation indexes for all the three diseases higher than 0.82 and those for liver fibrosis higher than 0.84. The results show that the proposed method may be a promising and competitive tool for the intelligent diagnosis of the three typical pathological forms of CLDs.