RESUMEN
OBJECTIVES: To determine the incremental diagnostic value of radiomics signature of pericoronary adipose tissue (PCAT) in addition to the coronary artery stenosis and plaque characters for detecting hemodynamic significant coronary artery disease (CAD) based on coronary computed tomography angiography (CCTA). METHODS: In a multicenter trial of 262 patients, CCTA and invasive coronary angiography were performed, with fractional flow reserve (FFR) in 306 vessels. A total of 13 conventional quantitative characteristics including plaque characteristics (N = 10) and epicardial adipose tissue characteristics (N = 3) were obtained. A total of 106 radiomics features depicting the phenotype of the PCAT surrounding the lesion were calculated. All data were randomly split into a training dataset (75%) and a testing dataset (25%). Then three models (including the conventional model, the PCAT radiomics model, and the combined model) were established in the training dataset using multivariate logistic regression algorithm based on the conventional quantitative features and the PCAT radiomics features after dimension reduction. RESULTS: A total of 124/306 vessels showed functional ischemia (FFR ≤ 0.80). The radiomics model performed better in discriminating ischemia from non-ischemia than the conventional model in both training (area under the receiver operating characteristic (ROC) curve (AUC): 0.770 vs 0.732, p < 0.05) and testing datasets (AUC: 0.740 vs 0.696, p < 0.05). The combined model showed significantly better discrimination than the conventional model in both training (AUC: 0.810 vs 0.732, p < 0.05) and testing datasets (AUC: 0.809 vs 0.696, p < 0.05). CONCLUSIONS: The PCAT radiomics model showed good performance in predicting myocardial ischemia. Addition of PCAT radiomics to lesion quantitative characteristics improves the predictive power of functionally relevant CAD. KEY POINTS: ⢠Based on the plaque characteristics and EAT characteristics, the conventional model showed poor performance in predicting myocardial ischemia. ⢠The PCAT radiomics model showed good prospect in predicting myocardial ischemia. ⢠When combining the radiomics signature with the conventional quantitative features (including plaque features and EAT features), it showed significantly better performance in predicting myocardial ischemia.
