Answers are grounded in this paper and cite their sources.
Hepatocellular carcinoma (HCC) remains a leading cause of global cancer-related mortality, driven by an ongoing etiological shift toward metabolic dysfunction-associated steatotic liver disease (MASLD). While early detection is vital for curative intervention, traditional surveillance and diagnostic modalities offer limited sensitivity due to operator dependence and inter-observer variability. This narrative review provides a comprehensive synthesis of emerging and investigational artificial intelligence (AI) applications across different stages of HCC management, ranging from research-stage predictive models to AI-assisted clinical tools. AI-driven radiomics workflows convert complex, non-invasive imaging profiles into high-dimensional data, enhancing the preoperative prediction of microvascular invasion (MVI) and refining standardized Liver Imaging Reporting and Data System (LI-RADS) classifications. Complementing radiology, computational pathology leverages deep learning and emerging pathological foundation models to standardize oncological scoring, mitigate inter-observer variability, and automate histological tumor segmentation. Multimodal AI fusion networks further bridge macroscopic imaging phenotypes with microscopic pathomic signatures and molecular profiles. At the population level, machine learning algorithms optimize dynamic surveillance by integrating longitudinal clinical data with liquid biopsy biomarkers, with the aim of detecting molecular change before it becomes visible on imaging. This review additionally examines AI integration within prospective clinical trial frameworks, encompassing endpoint standardization, patient stratification, and treatment response prediction. While technical challenges including scanner variability and retrospective validation gaps persist, explainable AI (XAI) and federated learning represent the next critical frontier for establishing transportable, accountable digital oncology workflows. Most current applications remain at the technical-validation stage and come largely from single-center, retrospective cohorts. Externally validated models are fewer, and evidence that AI improves clinical decisions or patient outcomes is still very limited.
No takes yet. Share an insight, caveat, or question.
Al‐Gizey et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: