An increasing number of studies have sought to identify ischemia by developing radiomics (Rad) models. Prior studies have demonstrated that computed tomography-based Rad features of peri-coronary adipose tissue (PCAT) exhibits satisfactory efficacy in identifying abnormal fractional flow reserve derived from coronary computed tomography (FFR CT ). However, its specific role in modeling remains unclear. This study aims to ascertain the impact of PCAT on the FFR CT , with a particular focus on the mediation analysis of the underlying mechanisms. In this cross-sectional study, vessels were divided into 2 groups: FFR CT ≤ 0.75 and FFR CT > 0.75. PCAT was semiautomatic outlined by 3 dimensions slicer and the Rad features were extracted. The dataset was divided into the train set (70%) and the test set (30%), and feature selection yielded the formula for Radiomics Score (Rad-Score). The role of Rad-Score in structural equation modeling was also determined by mediation analysis. Finally, Rad-Score models and Rad-Score + stenosis degree models were constructed using random forest (RF) methods. A total of 253 vessels from 174 patients were included. The mediation analysis demonstrated that the Rad-Score plays a significant role in mediating the impact of stenosis on ischemia, with a partial mediation effect (11.43% in left anterior descending artery, 6.38% in right coronary artery). The efficacy of Rad-Score combined stenosis to construct a RF model to identify FFR CT ≤ 0.75 was high, with an area under the curve of 0.879, sensitivity of 0.8, and specificity of 0.929 in the subgroup of left anterior descending artery; and an area under the curve of 0.840, sensitivity of 0.889, and specificity of 0.778 in the subgroup of right coronary artery. The proximal PCAT partially mediates the relationship between stenosis and coronary ischemia. The findings of the study indicate that RF models exhibit a high degree of identifying efficacy.
Chen et al. (Fri,) studied this question.