Arrhythmia ECG Beats Classification Using Wavelet-Based Features and Support Vector Machine Classifier

Introduction

In recent years, the incidence and prevalence of cardiovascular diseases (CVD) have gained an expanding growth. As per the recent report of WHO, the overall death rates due to CVD have decreased but the burden of heart diseases still remains at peak (Martis et al., 2013). Among many complications of CVD, atrial and ventricular arrhythmias are one of them which occur due to the abnormal electrical activity of heart muscles. Abnormal behavior of heart muscles generates different abnormal cardiac rhythms which are collectively known as arrhythmia. Ventricular fibrillation and flutter are some arrhythmia types which are life-threatening. It results in medical emergencies lead to cardiac arrest, hemodynamic collapse and sudden cardiac death (Huikuri et al., 2001).

The electrical activity of heart muscles could be monitored non-invasively using Electrocardiogram (ECG). Due to the small amplitude and duration, it is difficult to interpret the concealed information present in the ECG data. Therefore, a computer-assisted diagnostic tool could help the physicians in their practice and interpretation. The computer-assisted diagnosis tools can play a major role in the cardiovascular disease management (Hadhoud et al., 2006). These tools are based on many classifiers such neural network (Gutiérrez-Gnecchi et al., 2017), support vector machine (Übeyli, et al., 2007; Kolekar et al., 2015), deterministic learning (Dong et al., 2017), hidden Markov model (Liang et al., 2014; Dash et al., 2017), Bayesian network (de Oliveira et al. 2011; Kolekar et al., 2015; Kolekar et al., 2011), linear discriminant analysis (Yeh et al., 2009) etc. These classifiers are widely used for arrhythmia beat classification. A classifier based on the generalized linear model and the auto-regressive model was developed in (Ge et al., 2002) which claims 93.2% classification accuracy for six types of arrhythmia beats. A neuro-fuzzy technique with Hermite coefficients was used in (Linh et al., 2003) to classify ECG beats with 96% accuracy. In (Martis et. al, 2009), a Gaussian mixture model-based classification scheme was proposed for normal and abnormal ECG with 94% accuracy. An automated screening tool based on support vector machine classifier was proposed in (Martis et al., 2012) to distinguish normal and abnormal ECG beats with 95.6% accuracy.