Using Learnt Nakagami parametric mapping to classify fatty liver in rabbits

Abstract: Fatty liver disease is a condition where large vacuoles of fats accumulate in the liver cells. Based on statistical analysis of the backscattered signals, the ultrasound Nakagami image is an emerging technique for assessing FLD. In this study, we present a non-invasive deep learning method classifying liver steatosis based on ultrasonic Nakagami parametric mapping. Two different classifiers: convolutional neural network (CNN), the combination of CNN and multi-class support machine (MSVM), are used in this study. We use two sets of Nakagami parameters (constructed using the sliding window technique to process the raw backscattered fundamental and the second harmonic envelopes signal respectively) to train this classification model. Experimental results of 10-fold cross-validation show that the proposed improved classification model (CNN + MSVM with liner kernel) reaches a high classification accuracy (85.48%). Thus, such a method can be potentially beneficial to assist clinicians in diagnosis.