Cardiovascular disease continues to be the leading cause of death globally. Often, it stems from atherosclerosis, which can trigger substantial variations in the coronary arteries, possibly causing coronary artery disease (CAD). Coronary artery calcification is known to be a strong and independent forecaster of CAD. Hence, coronary computer tomography angiography (CTA) has become a fundamental noninvasive imaging tool to characterize coronary artery plaques. In this article, an automated algorithm is presented to uncover the presence of a calcified plaque, using 2060 CTA images acquired from 60 patients. Higher-order spectra cumulants were extracted from each image, thereby providing 2448 descriptive features per image. The features were then reduced using numerous well-established techniques, and ranked according to t value. Subsequently, the reduced features were input to several classifiers to achieve the best diagnostic accuracy with a minimum number of features. Optimal results were obtained using the support vector machine with a radial basis function, having 22 features obtained with the multiple factor analysis feature reduction algorithm. The accuracy, positive predictive value, sensitivity, and specificity obtained were 95.83%, 97.05%, 94.54%, and 97.13%, respectively. Based on these results, the technique could be useful to automatically and accurately identify calcified plaque evident in CTA images, and may therefore become an important tool to help reduce procedural costs and patient radiation dose.
K E Y W O R D Sautomated detection, CAD, CTA, higher-order spectrum cumulants