Artificial intelligence in hepatocellular carcinoma diagnosis and AI-assisted treatment support: Current landscape and future directions.
Authors
Affiliations (6)
Affiliations (6)
- Department of Medical Physics, Duke University, Durham, NC, USA.
- Institute of Regenerative Medicine, And School of Medicine, Jiangsu University, Zhenjiang, Jiangsu, China.
- Haihe Laboratory of Cell Ecosystem, Institute of Hematology, Chinese Academy of Medical Sciences, Tianjin, China.
- Guangdong Provincial Key Laboratory of Large Animal Models for Biomedicine, And South China Institute of Large Animal Models for Biomedicine, School of Pharmacy and Food Engineering, Wuyi University, Jiangmen, Guangdong, China.
- Department of Medical and Life Sciences, Faculty of Pharmaceutical Sciences, Tokyo University of Science, Tokyo, Japan.
- International Research Center for Medical Sciences, Kumamoto University, Kumamoto, Japan.
Abstract
Hepatocellular carcinoma (HCC) remains challenging to diagnose and manage, particularly at an early stage when curative treatment is most feasible. Artificial intelligence (AI) offers tools for integrating histopathology, radiology, laboratory testing, demographics, and electronic health records, while also creating opportunities for clinician-supervised AI-assisted procedural support. Deep learning applied to digitized pathology slides can support tumor subtype classification, grading, and microvascular invasion assessment. AI-driven imaging methods may support liver and lesion segmentation, small-nodule detection in cirrhotic livers, and non-invasive risk stratification when combined with radiomics and clinical variables. Emerging foundation models, large language models (LLMs), and agentic approaches may help structure evidence, summarize longitudinal records, and support guideline-aligned decision workflows. In AI-assisted treatment-support settings, robotic surgery, endoscopic platforms, and image-guided interventions should be viewed as environments for AI-assisted planning, navigation, documentation, and quality control rather than autonomous treatment systems. Most evidence remains retrospective, single-center, and vulnerable to domain shift. Other limitations include LLM/agentic reliability failures, limited interpretability, privacy constraints, computational burden, incomplete workflow integration, workforce readiness gaps, and uncertain clinical benefit. Future progress will require diverse multimodal datasets, prospective multicenter validation, explainable and lightweight algorithms, standardized data alignment, rigorous governance, patient-centered workflow implementation, staff training, and careful separation of technical feasibility from clinically meaningful outcomes. Responsibly deployed AI may support diagnostic consistency, workflow efficiency, and human-AI collaboration in HCC care.