Enhancing Liver Steatosis Diagnosis with Transfer Learning and Deep Convolutional Neural Network in Ultrasound Imaging

NAFLD is the most common liver abnormality, and liver biopsy is the gold standard for accurately quantifying liver steatosis in hepatic tissue samples. This study suggests an ultrasonic imaging-based neural network method for evaluating NAFLD. VGG16 deep CNN is pre-trained using the Image dataset in the proposed method, which enables extracting highlevel characteristics from liver B scan image sequences. It graded the degree of fatty liver in every liver using wedge-based biopsy. It compared our technique with the hepatorenal index method, a traditional CNN, and the GLCM methodology. Subsequently, the extraction of features classified the pictures as having fatty liver using the support vector machine (SVM) approach. Based on liver biopsies, fatty liver was defined as having over 5% of hepatocytes with steatosis. The Lasso regression approach was then used to combine the features and evaluate the amount of steatosis. The method proposed achieved a 0.977 area under the receiver operating characteristic curve (AUC). This method is effective and eliminates the need for ultrasound technicians to choose the region's interest manually, in contrast to earlier ones. Sonographers can use it to quickly identify the amount of fatty liver. The methodology's efficacy and precision indicate that the VGG16-based approach holds promise for extensive application in clinical contexts for the assessment of non-alcoholic fatty liver disease.