Epicardial adipose tissue radiomics from cardiac CT scans reveal distinct phenotypic patterns across cardiomyopathy subtypes

Introduction Epicardial adipose tissue (EAT) has been proposed as an imaging biomarker reflecting myocardial–adipose tissue interactions. Emerging evidence suggests that disease-specific alterations in EAT composition may be associated with distinct cardiomyopathy phenotypes, potentially mirroring underlying myocardial pathology. Purpose To explore whether radiomic features of EAT derived from routine cardiac CT scans are associated with distinct cardiomyopathy phenotypes, supporting the potential role of EAT radiomics as a non-invasive and reproducible imaging biomarker. Methods We retrospectively analyzed cardiac CT scans from 149 subjects: 30 with DCM/NDLVC/ACM, 15 with recent myocarditis (CT within 8 weeks of the event), 14 with HCM, 20 with ATTR-CM, 20 with aortic stenosis, and 50 controls. EAT was segmented using the open-source software 3D-Slicer, and 843 radiomic features were extracted using the PyRadiomics extension. Multigroup comparisons were conducted to assess radiomic feature differences across distinct cardiomyopathy phenotypes. Separate comparisons were performed using non-parametric testing with false discovery rate (FDR) correction: ATTR vs. Rest, DCM/NDLVC/ACM vs. controls, and myocarditis vs. controls. For each binary comparison, a composite radiomic score was constructed by averaging standardized, directionally aligned values of the most discriminative features. Diagnostic performance was assessed using receiver operating characteristic (ROC) analysis. Results EAT radiomics revealed distinct phenotypic patterns across cardiomyopathies. Multigroup heatmap visualization of top Kruskal-Wallis-ranked radiomic features revealed disease-specific texture patterns characterized by relative deviations from the cohort mean rather than uniform amplitude shifts (Figure 1A). Representative EAT radiomic features demonstrated distinct distributions across cardiomyopathy phenotypes, with statistically significant group-level differences (Figure 1B). Following feature-level analysis, the most informative EAT radiomic features for each binary comparison were selected and combined into a composite radiomic signature. These composite scores (Figure 2A, B, C) demonstrated good discriminatory ability, which was quantified using ROC analysis (Figure 2D, E, F), resulting in an AUC of 0.878 for ATTR-CM vs. rest, 0.812 for DCM/NDLVC/ACM vs. controls, and 0.884 for myocarditis vs. controls. The most informative features were predominantly wavelet-derived texture descriptors from GLCM, GLDM, and NGTDM families, indicating disease-specific alterations in EAT heterogeneity. Conclusion Radiomic analysis of epicardial adipose tissue on routine cardiac CT scans reveals distinct phenotypic patterns across major cardiomyopathy subtypes. These findings highlight the potential of EAT radiomics as an interpretable imaging biomarker for cardiomyopathy phenotyping and support further validation in larger prospective cohorts.Figure 1Figure 2