CT Radiomics of Pericoronary Fat to Predict Coronary Artery Calcification

Abstract

Background:Coronary artery disease (CAD) is a leading cause of mortality worldwide. Coronary artery calcification (CAC) is a key indicator of CAD, reflecting plaque burden. Pericoronary adipose tissue (PCAT) promotes vascular inflammation and contributes to plaque development, making it a promising imaging biomarker. This study aimed to create a radiomics-based model using cardiac CT features of PCAT around the left main coronary artery (LMCA) to predict CAC. Methods:Imaging from forty patients who underwent ECG-gated cardiac CT was retrospectively analyzed and grouped by CAC presence. Manual segmentation was performed using 3D Slicer to delineate PCAT surrounding the LMCA. Minimum Redundancy Maximum Relevance (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO) regression were utilized for feature selection. Random Forest and support vector machine (SVM) models were trained and compared. Results:From the 1,037 features extracted, two features with non-zero coefficients were retained at the optimal LASSO parameter (log α = 0.0128). The Random Forest model achieved 92% accuracy and an area under the curve (AUC) of 0.9143, outperforming SVM. Conclusion:Radiomic features of PCAT on cardiac CT can accurately predict CAC, showing its potential as an imaging-based biomarker for CAD risk assessment. J. Med. Invest. 72 : 330-336, August, 2025