Chest CT-based analysis of radiomic and volumetric differences in epicardial adipose tissue in HFrEF patients with and without AF.

Aims: Epicardial adipose tissue (EAT) has been implicated in atrial fibrillation (AF). While increased EAT volume (EATV) and EATV index (EATVI) are associated with AF, decreased values have been observed in heart failure with reduced ejection fraction (HFrEF). However, radiomic and volumetric differences of EAT in HFrEF patients with AF (HFrEF-AF) and without AF (HFrEF) remain unexplored.

Methods: This case-control study enrolled 120 patients (60 HFrEF and 60 HFrEF-AF). EATV and EATVI were quantified from non-contrast chest CT scans. Radiomic features were extracted using PyRadiomics, and reproducibility was assessed using intraclass correlation coefficients (ICCs). Feature selection was performed using the Boruta algorithm embedded in a five-fold cross-validation framework. Univariate and multiple logistic regression were used to explore group differences in echocardiographic parameters. Network correlation analysis and Mantel tests were conducted to examine associations between selected radiomic features and structural cardiac indices, such as left ventricular end-diastolic diameter (LVEDD) and right atrial volume index (RAVI).

Results: HFrEF-AF exhibited significantly higher EATV (174.22 ± 67.51 mL vs. 137.21 ± 58.03 mL; P = 0.002) and EATVI (102.68 ± 39.56 mL/m² vs. 80.60 ± 31.93 mL/m²; P = 0.001) compared to the HFrEF group. Significant differences were observed in 5 radiomic features between the groups. In the HFrEF subgroup, several radiomic features (e.g., original_shape_VoxelVolume and original_gldm_GrayLevelNonUniformity) correlated with LVEDD (Mantel test, 0.01 < P < 0.05, r > 0.05). In the HFrEF-AF subgroup, these radiomic features correlated with RAVI ( Mantel test, 0.01 < P < 0.05, r > 0.05).

Conclusions: HFrEF-AF exhibits distinct EAT imaging characteristics, including elevated EATVI and altered radiomic profiles, potentially reflecting cardiac structural changes. Notably, correlations between radiomic features and LVEDD were more pronounced in HFrEF, whereas associations with RAVI were prominent in HFrEF-AF, suggesting subgroup-specific structural patterns.