Diagnostic accuracy of artificial intelligence models for imaging detection of hepatic steatosis through systematic...

Non-alcoholic fatty liver disease (NAFLD) is an increasingly prevalent chronic liver condition affecting nearly 30% of the global population. Characterized by hepatic steatosis in the absence of significant alcohol intake, NAFLD can progress to non-alcoholic steatohepatitis (NASH), fibrosis, cirrhosis, and hepatocellular carcinoma. Although liver biopsy is the diagnostic gold standard, its invasiveness, cost, and associated risks limit widespread application. Artificial intelligence (AI) offers promising non-invasive alternatives by leveraging large datasets to enhance diagnostic precision. To evaluate the diagnostic accuracy of artificialintelligence algorithms for the imaging detection of hepatic steatosis, the essential first step in the metabolic dysfunctionassociated steatotic liver disease (MASLD) spectrum. A comprehensive literature search was conducted across PubMed, Scopus, Embase, Cochrane Library, and Google Scholar for studies published between January 2016 and January 2025. Studies were included if they involved adult populations, employed AI algorithms for NAFLD diagnosis, and reported sufficient diagnostic accuracy measures. Quality assessment was performed using the QUADAS-2 tool. Meta-analysis was conducted using a bivariate random-effects model to estimate pooled sensitivity, specificity, and area under the hierarchical summary receiver operating characteristic (HSROC) curve. Out of 29 studies included in the systematic review, 19 met the criteria for meta-analysis, comprising a total of 344,266 participants. AI-based diagnostic models showed excellent performance, with pooled sensitivity of 91% (95% CI: 84-95%), specificity of 92% (95% CI: 86-96%), and an AUC of 0.97 (95% CI: 0.95-0.98). The diagnostic odds ratio was 123.7, indicating high discriminatory capacity. Convolutional neural networks (CNNs) demonstrated superior accuracy (AUC = 1.00) compared to other AI classifiers. Subgroup analysis revealed higher diagnostic accuracy in studies validated with imaging standards compared to those using liver biopsy. Model performance was also influenced by the type of classifier and validation method used. AI-based models, particularly CNNs, exhibit high diagnostic accuracy for detecting hepatic steatosis and offer promising non-invasive alternatives to traditional modalities. These tools have the potential to transform early detection and screening, especially in resource-limited settings. Future research should focus on external validation, multicentric trials, and standardized reporting for clinical integration.