AI prediction of obstructive coronary artery disease using calcium-omics from non-contrast CT calcium scoring scans

Non-contrast computed tomography calcium scoring (CTCS) provides a cost-effective and accessible method for directly assessing coronary atherosclerosis. This study introduces an innovative machine learning framework designed for predicting obstructive coronary artery disease (CAD), categorized using the coronary artery disease-reporting and data system (CAD-RADS), based on CTCS scans. The analysis included 1,324 patients who underwent both CTCS and coronary CT angiography. Obstructive CAD was defined as CAD-RADS 4A-5, while CAD-RADS 0-3 were categorized as non-obstructive CAD. The study utilized clinical information, Agatston scores, and calcium-omics features to develop predictive models for obstructive CAD. Feature selection was performed using CatBoost and Shapley additive explanations methods, and the selected features were employed to train a CatBoost machine learning model. Model performance was assessed through 1,000 iterations of five-fold cross-validation. Among the participants, 334 patients (25.2%) had obstructive CAD. The top 11 predictive features consisted of 3 clinical variables, the Agatston score, and 7 calcium-omics features. The model achieved strong classification performance, with a sensitivity of 77.4±4.4%, specificity of 91.8±1.8%, and accuracy of 83.5±1.9%. Importantly, the inclusion of the Agatston score and calcium-omics features significantly improved predictive accuracy. These results highlight the potential of CTCS imaging in identifying individuals who may require further evaluation for obstructive CAD.