Pericoronary Adipose Tissue Radiomics-Based Prediction Models for Cardiovascular Events: A Systematic Review of...

Pericoronary adipose tissue (PCAT) radiomics provides a CT-derived phenotype of perivascular inflammation and tissue heterogeneity. This review evaluated the predictive performance, validation, incremental value, and methodological quality of PCAT radiomics-based prediction models for cardiovascular events.

This PROSPERO-registered systematic review (CRD420261417591) followed PRISMA 2020. PubMed and Web of Science Core Collection were searched for original human studies from 1 January 2014 to 8 June 2026. Data were extracted on populations, outcomes, region of interest definitions, radiomics workflows, model structure, performance, validation, calibration, clinical utility, and incremental value. Risk of bias and applicability were assessed using PROBAST.

Eighteen studies reported 74 prediction models, including 44 incorporating PCAT radiomics. Populations, outcomes, prediction horizons, region-of-interest strategies, and modelling workflows were heterogeneous, limiting comparison of AUCs and C-indices. Most outcomes were study-defined composite major adverse cardiovascular events; cardiovascular death and myocardial infarction were rarely evaluated separately. PCAT radiomics features were assessed as standalone signatures or incorporated into clinical-radiomics and multimodal models. Among studies with comparator models, most reported improved discrimination after adding PCAT radiomics, but formal incremental-value assessment was uncommon. Four studies (22.2%) reported external validation, external testing, or cohort-wide prognostic testing. Calibration was reported in 13 studies and decision-curve analysis in 15, but reporting was often incomplete or graphical. NRI or IDI was reported in 3 studies, and 7 assessed reproducibility or applied ICC-based feature filtering. PROBAST identified high overall risk of bias in 11 studies and unclear risk in 7; none had low overall risk.

The use of PCAT radiomics for cardiovascular event prediction warrants further investigation as a potential adjunct to CT-based risk stratification. However, methodological heterogeneity, limited external validation, incomplete incremental-value assessment, and risk of bias mean that improvement in routine risk stratification remains unestablished. Standardized workflows and external validation are required before implementation.