Analysis of epicardial adipose tissue in relation to arterial hypertension using radiomics in photon-counting CT

Cardiac CT is increasingly integrated into clinical routine. However, looking beyond the diagnostic aspect of identifying hemodynamically significant stenosis, individual risk stratification through interpretation of textural alterations of cardiac structures influenced by clinical risk factors, such as arterial hypertension, becomes increasingly relevant after the implementation of photon-counting CT. In this retrospective, single-center, IRB-approved study, 114 patients without coronary artery disease underwent PCCT. Epicardial adipose tissue (EAT) was manually segmented, and 1,015 radiomic features were extracted using PyRadiomics. Patients were divided into two groups defined by the presence or absence of arterial hypertension. Feature selection combined univariate analysis, recursive feature elimination, and random forest importance. Three machine-learning models were evaluated using a leakage-free protocol with repeated stratified cross-validation and a separately held-out test set. Logistic regression achieved the best but only moderate performance (cross-validated AUC 0.65; held-out test AUC 0.59). The most robust quantitative finding was a lower mean EAT attenuation in the hypertension group. These exploratory findings should be regarded as hypothesis-generating and require validation in larger cohorts.