Preoperative prediction of microvascular invasion in hepatocellular carcinoma: an umbrella review of systematic reviews and meta-analyses with etiology-specific implications

Background: Radiomics-based imaging models have been proposed for preoperative prediction of microvascular invasion (MVI) in hepatocellular carcinoma (HCC), but evidence is dispersed across multiple systematic reviews and meta-analyses and the influence of underlying liver disease etiology remains uncertain. This umbrella review aimed to summarise review-level diagnostic performance and assess etiology-related considerations.

Methods: An umbrella review of published systematic reviews and meta-analyses was conducted in accordance with the PRIOR statement. PubMed, Embase, and Web of Science were searched from inception to identify eligible reviews. Review-level data on imaging modality, modelling approach, and pooled diagnostic performance metrics were extracted. Evidence was synthesised descriptively, without additional quantitative pooling across reviews. Etiology-specific patterns were explored based on explicit reporting or region-based inference from included primary studies.

Results: Seven systematic reviews and meta-analyses were included. Radiomics-based imaging models demonstrated moderate-to-high diagnostic performance, with pooled area under the curve (AUC) values generally ranging from 0.80 to 0.90. Pooled AUCs overlapped substantially between studies dominated by hepatitis B virus (HBV)-endemic populations and those including mixed etiologic backgrounds, although several HBV-dominant reviews reported the highest AUC values. Ultrasound (US)-based radiomics showed lower diagnostic performance than computed tomography (CT) or magnetic resonance imaging (MRI)-based approaches. One review comparing radiomics-based and non-radiomics models reported comparable performance.

Conclusions: Radiomics-based models demonstrate generally favourable performance for predicting MVI in HCC, but methodological heterogeneity and inconsistent etiology reporting limit robust etiologic comparison and generalisability. Although higher diagnostic accuracy has most frequently been reported in HBV-endemic populations, these observations are exploratory and should be interpreted cautiously, as etiologic classification was partly inferred from regional patterns rather than consistently reported at the primary study level. Standardised etiology reporting, transparent modelling strategies, and external validation across diverse populations are required to clarify clinical utility.