Cardiac arrhythmia classification using tunable Q-wavelet transform based features and support vector machine classifier

Abstract

Electrocardiogram (ECG) is a non-invasive clinical tool that reveals the rhythm and functionality of the human heart. It is widely used in the diagnosis of heart diseases including arrhythmia. Abnormal heart rhythms are collectively known as arrhythmia which can be recognized and classified into different types. Arrhythmia classification techniques provide automated ECG analysis in cardiac patient monitoring devices. It helps cardiologists to interpret the ECG signal for diagnosis. In this context, this paper reports a novel and efficient ECG beats classification technique for normal and seven arrhythmia types. The proposed technique utilizes tunable Q-wavelet based features of ECG beats which are acquired from different ECG records of the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) arrhythmia database. For feature extraction, each ECG beat is decomposed up to the sixth level of the tunable Q-wavelet transform. Approximate coefficients at the sixth level are selected as features of each ECG beats. For classification, features of 14,878 ECG beats are utilized for training of the support vector machine classifier while 26,219 ECG beats are used for the testing purpose. The average accuracy, sensitivity, and specificity offered by the proposed classifier for eight different classes of ECG beats are 99.27%, 96.22%, and 99.58% respectively. The proposed classifier outperforms many recent techniques developed in this field.