JMIR Publications

ABSTRACT

Background:

The steadily rising global incidence of metabolic dysfunction-associated fatty liver disease (MAFLD) has garnered considerable concern within the international medical community. Although noninvasive imaging techniques offer the advantages of being painless and highly reproducible, they remain substantially limited in their ability to accurately quantify the degree of hepatic steatosis (HS). Therefore, histopathological examination of liver tissue remains the “gold standard” for diagnosis. The fast advancement of artificial intelligence (AI) technologies is reshaping paradigms in medical image analysis and presents fresh opportunities for the intelligent HS identification and grading.

Objective:

Our study endeavors to evaluate the diagnosis performance and clinical utility of AI-enabled models in HS assessment via a comprehensive systematic review and meta-analysis, thereby providing evidence-based support for their potential clinical translation.

Methods:

PubMed, Cochrane Library, Embase, Web of Science, and IEEE were thoroughly retrieved to screen out eligible studies. Extracted outcomes were the area under the curve (AUC) with corresponding 95% confidence intervals (CIs), sensitivity, specificity, diagnostic accuracy, positive and negative predictive values, as well as false positive/negative rates. Data from qualified studies were synthesized for our meta-analysis.

Results:

32 studies were eligible for systematic review, of which 29 were eligible for quantitative synthesis. The pooled sensitivity was 95% (95% CI: 92–96%), specificity was 92% (95% CI: 90–94%), and the pooled AUC was 0.92 (95% CI: 0.90–0.94). Subgroup analyses were conducted based on the type of AI employed (machine learning (ML) vs deep learning (DL)), reference standard (magnetic resonance imaging (MRI)-proton density fat fraction, liver histopathology, ultrasonography (US)), imaging modality (US, computed tomography(CT), histology), and application of transfer learning (TL).

Conclusions:

This meta-analysis demonstrates the high diagnostic efficacy of AI in HS evaluation. Nevertheless, the clinical implementation of such technologies remains constrained by challenges related to data quality, model architecture, and validation frameworks. With continued refinement of algorithmic design, standardization of data acquisition protocols, and promotion of interdisciplinary collaboration, AI holds considerable promise as a pivotal tool for precise metabolic liver disease diagnosis and management, ultimately contributing to the alleviation of the growing global burden of hepatic disorders. Clinical Trial: PROSPERO number: CRD420251046862