Predicting OCT-defined vulnerable plaques and MACE using quantitative imaging features of pericoronary adipose tissue from coronary CTA

OBJECTIVE Pericoronary adipose tissue (PCAT) attenuation detected on coronary computed tomography angiography (CCTA) has been associated with coronary inflammation and cardiovascular risk. We aimed to construct PCAT-based deep learning (DL) radiomics models for predicting plaque vulnerability assessed by optical coherence tomography (OCT) and major adverse cardiovascular events (MACE). METHODS A total of 116 patients (training set, n = 92; validation set, n = 24) with coronary artery disease (CAD) who underwent CCTA and OCT within three months between April 13, 2019 and February 23, 2023 were included. OCT identified vulnerable-plaque profiles, including thin cap fibrous atheroma (TCFA), macrophage, and cholesterol crystal. PCAT was segmented around the target vessel segment of the coronary culprit vessel. Radiomic and deep features (DF) of PCTA were extracted using PyRadiomics and a simple-yet-effective Volume Contrast framework, respectively. Principal component analysis and least absolute shrinkage and selection operator were leveraged to select the features. The clinical model was built using logistic regression analysis, while the clinical-radiomics model and clinical-radiomics-DF model was developed using support vector machine. RESULTS In the validation cohort, the clinical-radiomics-DF model outperformed the clinical and the clinical-radiomics models. The AUC values were as follows: 0.708 for TCFA compared to 0.625 and 0.687; 0.738 for macrophage infiltration compared to 0.575 and 0.833; 0.771 for cholesterol crystals compared to 0.542 and 0.721; and 0.778 for MACE compared to 0.698 and 0.746. CONCLUSION A noninvasive DL radiomics model based on PCAT features may facilitate the identification of vulnerable plaques and CAD patients at high risk for future adverse events.