A Lesion-Specific Coronary Artery Calcium Quantification Framework for the Prediction of Cardiac Events

Abstract

CT-based coronary artery calcium (CAC) scanning has been introduced as a noninvasive, low-radiation imaging technique for the assessment of the overall coronary arterial atherosclerotic burden. A 3-D CAC volume contains significant clinically relevant information, which is unused by conventional whole-heart CAC quantification methods. In this paper, we have developed a novel distance-weighted lesion-specific CAC quantification framework that predicts cardiac events better than the conventional whole-heart CAC measures. This framework consists of 1) a novel lesion-specific CAC quantification tool that measures each calcific lesion's attenuation, morphologic and geometric statistics; 2) a distance-weighted event risk model to estimate the risk probability caused by each lesion; and 3) a Naive Bayesian-based technique for risk integration. We have tested our lesion-specific event predictor on 60 CAC positive scans (20 with events and 40 without events), and compared it with conventional whole-heart CAC scores. Experimental results showed that our novel approach significantly improves the predictive accuracy, indicated by an improved area under the curve of receiver operating characteristic analysis from 62% to 68%, an improved specificity by 23-55% at the 80% sensitivity level, and a net reclassification improvement of 30% .