Non-destructive detection of CAD stenosis severity using ECG-PCG coupling analysis

Abstract

Background and ObjectiveEffectively assessing the coronary stenosis degree is vital for the diagnosis of coronary artery disease (CAD). Electrocardiogram (ECG) and phonocardiogram (PCG) are simple, affordable, and non-invasive approaches, containing large cardiovascular physio-pathology information. This study aims to develop a non-destructive technique for CAD stenosis severity assessment. MethodsBy extracting the time series data from ECG and PCG signals, three coupling feature types, namely, cross-entropy (XEntropy), non-entropy (NEntropy), and multivariate-entropy features (MEntropy), are constructed. Then, a statistical analysis is conducted. Combing recursive feature elimination and support vector machines, coronary artery stenosis severity is identified. Experiments are performed on 237 simultaneously obtained ECG and PCG signals from the severe CAD (sCAD), mild to moderate CAD (mCAD), suspected CAD (CPNCA), and healthy (Health) subject groups. ResultsStatistical analysis results show that numerous features are statistically different. Regarding single-class coupling features, MEntropy has the strongest detection ability, followed by XEntropy, while that of NEntropy is relatively poor. Combining these three types of coupling features further improved the accuracy; the accuracies of sCAD-mCAD, sCAD-CPNCA, mCAD-CPNCA, sCAD-Health, mCAD-Health, and CPNCA-Health were 91.17%, 96.64%, 93.33%, 99.39%, 98.89%, and 97.78%, respectively. ConclusionsOur study confirms the great potential of MEntropy in characterizing cardiovascular status and provides a feasible solution for non-destructively assessing coronary artery stenosis.