Harnessing AI-driven approaches for detecting metabolic dysfunction-associated steatotic liver disease, assessing fibrosis, and stratifying hepatocellular carcinoma risk: a scoping review.

INTRODUCTION

Non-alcoholic fatty liver disease (NAFLD) is the most prevalent chronic liver disorder worldwide and a major risk factor for hepatocellular carcinoma (HCC). Its rising prevalence and progression to HCC present major clinical and public health challenges. Current diagnostic and prognostic tools lack accuracy in predicting disease progression and early HCC risk. Emerging approaches, including artificial intelligence (AI), machine learning (ML), and polygenic risk scores (PRS) offer promising opportunities and improvements in non-invasive diagnosis, risk stratification, and precision medicine.

METHODS

We conducted a scoping review synthesising evidence on the application of PRS and AI/ML models for predicting and detecting MASLD, assessing fibrosis, and stratifying risk of HCC among individuals with MASLD/NAFLD, with a particular focus on European and Asian populations. The review was performed in accordance with PRISMA 2020 guidelines and aimed to map the development trajectory and knowledge structure of predictive approaches for MASLD/NAFLD-HCC risk stratification, while identifying key translational challenges and opportunities.

RESULTS

Evidence from the included studies indicates a shift from isolated methodological innovation towards integrated, explainable, and clinically validated multimodal models, supported by transparent AI/ML systems aligned with regulatory frameworks. There is also a critical need for large, multi-centre validation studies and interdisciplinary collaboration among clinicians, data scientists, and policymakers to enable scalable and equitable implementation. Furthermore, progress depends on moving from theoretical promise to clinical reality through prospective trials that assess real-world effectiveness and adoption pathways.

CONCLUSION

This review highlights advances in AI and PRS for HCC risk prediction in MASLD/NAFLD, demonstrating that multimodal models outperform traditional approaches. AI-driven methods show strong potential for MASLD detection, fibrosis assessment, and HCC risk stratification, with implications for improved clinical outcomes. However, translation into practice is limited by poor genetic integration, lack of validation, population bias, and limited explainability. Further validation, standardization, and clinical integration are required for widespread adoption and effective personalised surveillance.