Wavelet Packet Analysis of ECG signals to Understand the Effect of a Motivating Song on Heart of Indian Male Volunteers

Introduction

Music has been reported to have an advantageous impression on the human health (MacDonald, Kreutz, & Mitchell, 2013). It has been reported to arouse the emotions (Juslin & Sloboda, 2001) and can be used as a stimulus for triggering the emotional responses for the person. Recent studies have suggested that listening to the music is also associated with a change in the physical states (e.g., promotes sleep, reduces anxiety, increases positive attitude towards exercise, and increases brain activities) of the person (Lai, 2001) . Therefore, it can be told that a temporary change occurs in both the physiological and the psychological states of the person while listening to the music (Levenson, 2014; Yen et al., 2010). Apart from exploring the above-mentioned physiological effects of the music, research is going on to establish the effect of music on the heart. The electrical activity of the heart can be analyzed non-invasively using electrocardiogram (ECG) signals, obtained from electrocardiography test. Electrocardiography is a painless and uncomplicated test (Clark, 2015; De Mello, 2013). It records the electrical activity of the heart as a function of time through the placement of the electrodes on the skin. As the ECG signal is acquired in a non-invasive condition, its classification has been extensively studied by the researchers for the diagnosis of the cardiac diseases as well as for identifying any alteration in the cardiac electrophysiology due to a stimulus (Yaman et al., 2013). The classification of the ECG signals usually comprises of four stages, i.e., preprocessing, segmentation, feature extraction, and classification. The pre-processing level typically performs artefact removal, signal enhancement and normalization. Then, segmentation fragments the signal into smaller segments that indicate the electrical activity of the heart in a better way (Da S Luz, Nunes, De Albuquerque, Papa, & Menotti, 2013). From these two stages, good results can be obtained using some prevalent features (Li & Zhou, 2016; Da S Luz, Merschmann, Menotti, & Moreira, 2017). Feature extraction is essential for the ECG signal classification and can be performed using the raw or segmented ECG signals. Various types of methods have been proposed for the extraction of the ECG signal features, such as, the extraction of the sample points at some frequency from the ECG signal, statistical and/or morphological methods, and transformed domain methods. Amongst the transformed domain methods, wavelet-based methods such as discrete wavelet transform (DWT) and its extension to wavelet packet decomposition (WPD) have gained significance because of their ability to capture the centralized spatial-frequency information of the ECG signals (Li & Zhou, 2016). However, WPD has been reported to provide better results than DWT. This may be because of the fact that it decomposes both the approximation and the detail coefficients of the ECG signals. Over the years, classification of the ECG signals has been attempted by numerous methods such as support vector machine (SVM), K-nearest neighbour (KNN) and ANN in different studies. Amongst the supervised learning-based classification methods, ANN has received special attention of the researchers because of the multi-parametric nature of its architecture (Alexakis et al., 2003). Taking the motivation from the above-mentioned discussion, the current study proposes the use of WPD analysis and ANN classification of ECG signals to understand the effect of a motivational song on the cardiac electrophysiology.