Discrimination of Life-Threatening Arrhythmias Using Singular Value, Harmonic Phase Distribution, and Dynamic Time Warping of ECG Signals

The automated analysis of electrocardiogram (ECG) signal is crucial for early recognition of life-threatening arrhythmias. This work presents an efficient ECG feature description model for diagnosis of ventricular arrhythmias (VAs); ventricular fibrillation (VF) and ventricular tachycardia (VT). The concept of a new feature set by applying singular value decomposition (SVD) and harmonic phase distribution is introduced to capture the morphological variation of the ECG signal. The phase characterization and singular value features are extracted from discrete Fourier transform and Empirical mode decomposition of ECG signal, respectively. In addition to that, dynamic time warping (DTW) procedure is applied to measure the similarity/dissimilarity pattern of ECG signals for classifying VAs. The firstfold classification involves the discrimination of VAs and non-VA patterns. Following this, secondfold classification discriminates VF versus VT and normal rhythms from other arrhythmic conditions. Experimental results are validated using the benchmark Physionet ECG database. The proposed method achieved the best performances using 5-s ECG segment with tenfold cross validation strategy. A comparative study with existing methods also shows superior performances of the proposed feature extraction scheme.