Non-Alcoholic Hepatic Steatosis Grading: Exploring the AI Potential of Machine vs Deep Learning

Abstract—Non-alcoholic fatty liver disease(NAFLD)is a grow- ing global health concern and a leading cause of chronic liver disease. Accurate, early-stage detection and grading of hepatic steatosis are essential to prevent progression to more severe con- ditions such as cirrhosis or hepatocellular carcinoma. This study evaluates the performance of classical machine learning and deep learning approaches for automated classification of liver steatosis using ultrasound (US) images from the BEHSOF dataset, which includes annotated clinical metadata. Preprocessing techniques such as grayscale normalization and Local Binary Pattern (LBP) feature extraction were employed to enhance diagnostic features. Three models—Support Vector Machine(SVM), Random Forest (RF), and an Artificial Neural Network (ANN)—were developed and tested. The ANN achieved the highest accuracy (99.09%), outperforming both RF(99.0%)and SVM(80%)classifiers, and surpassing previously reported Inception-ResNet-v2 results. These findings highlight the potential of interpretable and computationally efficient AI systems in supporting ultrasound- based NAFLD assessment, especially