Medical knowledge along with electrocardiogram (ECG) arrhythmia classification using artificial intelligence-based methods can be very useful and effective to treat a patient. ECG arrhythmia classification remains a challenging problem due to the need for a detailed analysis of the characteristics extracted from an ECG signal. Therefore, addressing this diagnostic field using signal processing techniques can be very valuable. In this paper, a combination of wavelet packet transform features, morphological coefficients, and the features extracted from the empirical mode decomposition technique is used to determine the ECG arrhythmia type. A sparse dictionary learning procedure based on a coherence criterion is proposed to learn the combinational feature vector related to the different arrhythmia classes. The proposed method minimizes the mutual coherence between the atoms of each dictionary related to the different arrhythmia categories. The proposed method is compared with other classification algorithms that employ different statistical and morphological features. The results show that the proposed algorithm can precisely classify the ECG arrhythmia types.
