Abstract
BACKGROUND: The global rise of metabolic associated fatty liver disease reflects the urgent need for accurate, noninvasive diagnostic approaches. The invasive nature of liver biopsy and the limited sensitivity of ultrasound in detecting early steatosis highlight a critical diagnostic gap. Artificial intelligence (AI) has emerged as a transformative tool, enabling the automated detection and grading of hepatic steatosis (HS) from medical imaging data. OBJECTIVE: This review aims to quantitatively evaluate the diagnostic performance of AI models for HS, explore sources of interstudy heterogeneity, and provide an appraisal of their clinical applicability, translational potential, and the major barriers impeding widespread implementation. METHODS: PubMed, Cochrane Library, Embase, Web of Science, and IEEE Xplore databases were searched until September 24, 2025. Studies using AI for HS diagnosis, meeting predefined PIRT (Patient Selection, Index Test, Reference Standard, Flow and Timing) framework and providing extractable data were included. Diagnostic performance indicators, including sensitivity, specificity, and the area under the summary receiver operating characteristic curve (AUC), were extracted and quantitatively synthesized. Meta-analyses were conducted using a bivariate random effects model. The methodological quality and risk of bias were evaluated using the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2) tool. Heterogeneity was assessed through the I² statistic, bivariate box plots, 95% PIs, and threshold effect analysis. Clinical applicability was examined using the Fagan nomogram and likelihood ratio tests. RESULTS: A total of 36 eligible studies were identified, of which 33 (comprising 36 cohorts) were included in the subgroup analyses. Results demonstrated excellent diagnostic accuracy of AI models, with a summary sensitivity of 0.95 (95% CI 0.93-0.96), specificity of 0.93 (95% CI 0.91-0.94), and an AUC of 0.98 (95% CI 0.96-0.99). Clinical applicability analysis (positive likelihood ratio >10; negative likelihood ratio <0.1) supported AI's strong potential for both confirming and excluding HS. However, substantial heterogeneity was observed across studies (I² >75%). According to QUADAS-2, a high risk of bias, particularly in the Patient Selection domain (44.4%), may have contributed to the overestimation of real-world performance. Subgroup analyses showed that deep learning models significantly outperformed traditional machine learning approaches (AUC: 0.98 vs 0.94). Models using ultrasound or histopathology references, retrospective designs, transfer learning, and public datasets achieved the highest accuracy (AUC 0.98-0.99) but contributed to interstudy heterogeneity. CONCLUSIONS: AI demonstrates remarkable potential for noninvasive screening and assessment of HS, especially in primary care. Nonetheless, clinical translation remains limited by performance variability, retrospective designs, lack of external validation, practical barriers such as data privacy and workflow integration. Future studies should prioritize prospective multicenter trials and standardized external validation to bridge the gap between current evidence and clinical application. The key innovation of this review lies in establishing a unified, modality-agnostic analytical framework that integrates evidence beyond single-modality evaluations.