Artificial Intelligence-Assisted Tools in the Diagnosis of Liver Infections: A Narrative Review

Liver infections, caused by bacterial, parasitic and viral agents, remain a major cause of global morbidity and mortality. They are diagnostically challenging because their clinical and imaging features overlap and specialist expertise is unevenly distributed. This narrative review synthesizes current evidence on artificial intelligence (AI)-assisted tools for the diagnosis, staging and risk stratification of liver infections. Machine learning, deep learning, radiomics and natural language processing have been applied across various liver infections, with multimodal models consistently outperforming single-modality approaches. Deep learning applied to ultrasound and contrast-enhanced CT achieved the strongest results, most notably in hepatic echinococcosis, while radiomics and machine learning models performed well for pathogen classification in pyogenic abscess and for fibrosis staging in schistosomiasis and hepatitis B. Evidence on AI application for amoebic liver abscess, fasciolosis, hepatic tuberculosis or cytomegalovirus hepatitis is nearly absent. Current evidence therefore supports technical feasibility rather than demonstrate clinical benefit. Major limitations impeding clinical application of AI tools include data scarcity and class imbalance, limited interpretability, retrospective single-centre designs, sparse external validation and inconsistent reporting of detailed performance metrics. Priorities include multicenter datasets from endemic, resource-limited settings, prospective validation, explainability and health-economic evaluation. Importantly, AI should be a reliable tool that support rather than replace clinician judgement.