Article Preview
TopIntroduction
The working principle of the tissues and organs in our body is based on the potential difference that occurs as a result of the electrochemical events of the cells. This potential difference produces electrical signals that can be measured from the body surface. The electrical activity of the heart is also measured and evaluated by the electrocardiography (ECG) method. ECG is the recording of the potential difference that occurs due to the contraction and relaxation of the heart during a heartbeat with the help of electrodes placed on the body surface. In a healthy person’s ECG signal, there are P waves, QRS complexes, and T waves, each representing different phases of the heartbeat. Analysis and interpretation of ECG signals recorded for a certain period of time play an important role in the diagnosis of any heart-related disease. Abnormalities caused by the wave formation time, shape or the time difference between the waves cause arrhythmic heart rhythms (Xiong et al., 2018). Early and accurate diagnosis of arrhythmic signals is critical in preventing diseases that may result in sudden death.
Manual analysis of the ECG signal is not sufficient for rapid detection of abnormalities in heart rhythm. The analysis of the long-term ECG signal by the experts takes a lot of time and this analysis may not accurately identify the problem. Computer-aided decision systems are being developed in the examination of ECG signals due to its advantages such as increasing the accuracy of diagnosis, shortening the analysis time, and reducing the expert errors that may occur. In the literature, there are many different studies based on signal processing methods related to arrhythmia detection using ECG signals. These studies are based on the extraction of different features from the signals and the classification of these features.
Depending on the examination of the time or frequency domain, ECG morphology (Anwar et al., 2018), RR interval (Xiang et al., 2018), principal component analysis (PCA) (Manik et al., 2019), independent component analysis (ICA) (Desai et al., 2015), Fourier transform (Kurniawan et al., 2020), empirical mode decomposition (EMD) (Izci et al., 2018), The features are extracted by methods such as discrete wavelet transform (DWT) (Hamed & Owis, 2016; Chen & Maharatna, 2020). Different machine learning algorithms are used in the classification of feature vectors obtained from arrhythmic and normal signals. Support vector machines (SVM) (Li et al., 2020), k-nearest neighbor (k-NN) classifier (Qaisar et al., 2020), artificial neural networks (ANN) (Dewangan & Shukla, 2016; Subbiah & Subramanian, 2018) are among the classifiers used in the classification of ECG signals. In traditional machine learning algorithms, the signals go through the preprocessing stage in order to decompose the noise that may occur during signal recording. At this stage, the signal is cleared of noise by various filters. Deep learning algorithms are being developed as an alternative to machine learning algorithms consisting of preprocessing, feature extraction and classification stages. Deep learning can perform preprocessing, feature extraction and classification stages together, thanks to many hidden layers in its structure (Hatami et al., 2018).
In this paper contribution are as follows,
An Alexnet-NCA-SVM hybrid structure is suggested for Arrhythmia, CHF, and NSR ECG signal categorization. The proposed method is implemented with hybridization of time frequency scalogram images to get deep features, Morphological features, DTCWT and PCA features fused with NCA to demonstrate its superiority in classifier. This paper's contribution to the literature can be summarized as follows. ECG signals for arrhythmia, CHF, and NSR are grouped together. The Alexnet-NCA-SVM hybrid structure was used for classification.