Review: Multi-lead Discrete Wavelet-based ECG Arrhythmia Recognition via Sequential Particle Support Vector Machine Classifiers

This study proposes a sequential heart arrhythmia classification method based on the analysis of multi-lead discrete wavelet transform (DWT)-derived metrics. First, a trous DWT of the baseline wander removed and scaled multi-lead electrocardiogram signal is computed. Scales of 2(2), 2(3), and 2(4) and a smoothing function scale of 2(2) are selected. Then, using a simple approximation technique, the fiducial and J points of each detected QRS complex are estimated and then the complex and the corresponding chosen DWT scales are segmented. Next, for each excerpted segment, the second- (variance), third- (quasi-skewness), and fourth- (quasi-kurtosis) order statistical moments and the curve length (as a nonlinear moment) are calculated and used as elements of the feature vector (20 measures for the feature vector of each lead). The proposed features are used for regulating the parameters of five sequentially operating particle radial basis function (RBF)-based support vector machine (SVM) classifiers implemented for multi-lead records of the MIT-BIH Arrhythmia Database for atrial premature (AP), nodal (junctional) premature (NP), ventricular escape (VE), nodal (junctional) escape (NE), abberated atrial premature (AAP), rhythm-changed (RC), ventricular-normal fusion (VNF), paced-normal fusion (PNF), non-conducted P-wave (NCPW), ventricular flutter (VF), paced beat (PB), premature ventricular contraction (PVC), left bundle branch block (LBBB), right bundle branch block (RBBB), and Normal categories. To increase the accuracy of the classification algorithm, the training database is divided into five groups, namely Normal+LBBB+RBBB+APB+PB (Group#0), NP+VE+NE (Group#1), AAP+RC+PVC (Group#2), VNF+PNF (Group#3), and VF+NCPW (Group#4). An RBF-SVM classifier is tuned for each group. In the proposed sequential particle classification algorithm, beats belonging to Group#0, Group#1, Group#2, and Group#3 are recognized and isolated from the test database via Classifier#0, Classifier#1, Classifier#2, and Classifier#3, respectively. Then, the arrhythmias of Group#4 are identified by Classifier#4. The proposed heart arrhythmia classification is applied to categorize the above mentioned arrhythmias. Average values of Se = 99.34%, P+ = 99.60%, Sp = 99.63%, and Acc = 98.64% are obtained for sensitivity, positive predictivity, specificity, and accuracy, respectively, showing marginal improvement of the heart arrhythmia classification performance. The proposed algorithm has acceptable performance, a low training computational burden, and a quick response.