Multi-Class Deep Learning Approaches for Automated Liver Disease Diagnosis: A Review

Liver diseases constitute a major global health burden, ranging from hepatitis and fatty liver to fibrosis, cirrhosis, and hepatocellular carcinoma. Early and accurate detection of disease stages is critical for effective clinical intervention, yet conventional diagnostic approaches often rely on invasive biopsies, expert-driven imaging interpretation, or isolated biomarker thresholds. Recent advances in deep learning, particularly multi-class deep neural networks, have enabled automated, scalable, and data-driven liver disease detection using heterogeneous clinical data. This review presents a comprehensive analysis of multi-class deep neural network–based methods for liver disease detection, integrating laboratory biomarkers, ultrasound, CT, MRI, and multi-modal clinical data. The study synthesizes recent progress in neural architectures, feature fusion strategies, explainable AI techniques, and weakly supervised learning frameworks. Emphasis is placed on how multi-class formulations improve disease stratification compared to binary classifiers and enhance clinical decision support. Furthermore, this review identifies key research findings, methodological strengths, and persistent challenges such as data imbalance, interpretability, and real-world deployment. By consolidating current evidence, the paper aims to guide researchers toward robust, interpretable, and clinically viable deep learning solutions for liver disease detection.