Artificial intelligence imaging for metabolic dysfunction-associated steatotic liver disease: Systematic review, meta-analysis, and treatment-response evidence assessment

This is the authors' abstract. We don't add key points for this paper.

BACKGROUNDImaging artificial intelligence (AI) has been studied mainly for cross-sectional steatosis and fibrosis assessment in metabolic dysfunction-associated steatotic liver disease (MASLD); its role in serial pharmacological-response assessment is less clear.AIM To synthesize imaging AI evidence for diagnosis, staging, and longitudinal treatmentresponse assessment in MASLD.METHODS PubMed and Scopus were searched on August 26, 2026, for reports available through June 30, 2026.One reviewer screened records, extracted data, and applied the Quality Assessment of Diagnostic Accuracy Studies 2; the protocol was not prospectively registered.Only patient imaging or imaging-signal models were eligible.Compatible area-under-the-curve estimates were pooled with logit-scale restricted-maximumlikelihood random effects and a Hartung-Knapp interval.RESULTS After screening 1011 unique database records and 18 additional reports, 41 reports were included.Only three independent cohorts (n = 356) shared the same patient-level target: Ultrasound AI for magnetic resonance imaging (MRI) proton density fat fraction ≥ 5% steatosis.Their exploratory pooled area under the curve was 0.94 (95% confidence interval: 0.38-1.00;I 2 = 85.3%); the prediction interval spanned approximately 0-1.Fibrosis evidence was not pooled.One 145-patient trial analysis assessed 12-week treatment-related MRI changes, but used internal cross-validation and did not validate an individual-response endpoint externally.4 / 59 CONCLUSION Cross-sectional discrimination can be high, but heterogeneity and limited validation prevent general performance claims.Longitudinal evidence remains a single exploratory study.