Concurrent Analysis of Dynamic and Static Features for Classifying Cardiac Rhythms

Arrhythmias confer a substantial risk of sudden cardiac death, accounting for over 80% of cases. The standard 12-lead electrocardiogram (ECG) offers a painless and noninvasive way of diagnosing cardiac arrhythmias. Considering the complementary value of dynamic and static features of the ECG, this study concurrently analyzes both aspects to comprehensively describe arrhythmia patterns in ECGs for enhancing performance. First, we extract new intrinsic dynamic features of ECG that characterize ECG temporal variations through dynamic learning from the perspective of nonlinear dynamics. Then, we consider not only these dynamic features but also static features, such as QT interval and analyze the significance of these features in classification through feature importance. Furthermore, based on the classification results of random forests (RFs), posterior probabilities are calculated and used to construct the basic probability assignment (BPA) in Dempster–Shafer (DS) evidence theory, thereby achieving the fusion of dynamic and static features. Finally, the method of this article is validated on the largest and publicly available Champon arrhythmia database. The findings indicate that the model employing the proposed dynamic features achieves an accuracy of 91.71% and 88.93% for identifying four and seven rhythms. Significantly, when analyzing dynamic and static features concurrently using DS theory, there is a considerable enhancement in the accuracy for detecting the four and seven rhythms, with an increase of 4.59% and 5.26%, respectively. These results illustrate the high potential value of the proposed arrhythmia detection method through the concurrent analysis of dynamic and static features.