A Deep Learning Architectures for Liver Disease Detection: From CNNs to Transformers and Beyond

Metabolic Dysfunction-Associated Fatty Liver Disease (MAFLD), previously referred to as Non-Alcoholic Fatty Liver Disease (NAFLD), is currently established as the world’s most prevalent chronic liver disease. MAFLD impacts almost 25 to 30% of the world’s population. The condition has the potential to develop into extreme conditions such as nonalcoholic steatohepatitis (NASH), fibrosis, cirrhosis, and hepatocellular carcinoma. MAFLD is among the primary causes of liver transplantation globally, accounting for approximately 20 to 25% of liver transplants in the United States. Its mortality rate also keeps growing, as liver-related complications and cardiovascular disease are the most common causes of death. Transformers and Convolutional Neural Network (CNN), among other trending model architectures are applied in deep learning for image classification and recognition activities. This article is a comparative review of CNNs, Vision Transformers (ViTs), and other mentioned detection models for automated histological classification and localization of such conditions. In the course of this review, we seek to bridge deep learning technology with applied clinical uses. We emphasize how models such as CNNs, Transformers, have the potential to greatly enhance liver disease detection and classification at the cellular level.