Ultrasound Image-Based Liver Disease Diagnosis Using Machine Learning

Abstract

A dataset comprising liver B-mode ultrasound image sequences is processed to construct a system to automatically detect non-alcoholic fatty liver disease (NAFLD) - also called metabolic dysfunction-associated steatotic liver disease (MASLD) - via machine learning-based binary classification. Owing to the paucity of samples in the dataset, alternatives to the widely used convolutional neural network (CNN) and similar deep learning approaches are adopted since harnessing CNN and comparative systems is likely to result in overfitting of the data. Simpler alternative machine learning approaches such as random forest classifier, logistic regression and decision tree classifier are employed, compared and contrasted. For the minimal datasets utilized, these simpler approaches resulted in reasonable performance metrics. Further refinement of the techniques could permit improvements in robustness and performance that could warrant incorporation of the resulting machine learning models into modules for the automated detection of liver disease based on ultrasound image sequences in a comprehensive artificial intelligence-driven healthcare system.