Cardiovascular disease (CVD) is a broad term encompassing a group of heart and blood vessel abnormalities that is the leading cause of death worldwide. The most popular and low-cost diagnostic tool for assessing the heart electrical impulses is an electrocardiogram (ECG). Automation is required to reduce errors and human burden while interpreting ECG signals. In recent years, deep learning shows better performance in ECG classification and has also shown that automated classification of ECG signals can improve accuracy and efficiency. In this chapter, the authors review the research work on ECG signals using deep learning methods like deep belief network (DBNK), convolutional neural network (CNNK), long short-term memory (LSTMY), recurrent neural network (RNNK), and gated recurrent unit (GRUT). In the research articles published between 2017 and 2021, CNNK was found to be the most appropriate technique for feature extraction.
TopIntroduction
Cardio-Vascular Disease (CVD) is a broad term that refers to illnesses that affect the heart and blood arteries of a person's body. Damage to the arteries in organs such as the heart, brain, eyes, and kidneys is also a possibility. CVDs can cause blood vessel blockage and blood clot development, which can result in either cerebral or cardiac ischemia, necrosis resulting in myocardial infarction (MI). Every part present in the body may turn out to be congested and starved of oxygen as a result of the heart's long-term inefficient blood pumping, resulting in varying degrees of damage (Yu et al. 2015). Even among young people, CVD is one of the leading causes of death in many established and developing countries around the world. However, adopting a healthy lifestyle can greatly reduce the risk of developing it.
Electrocardiogram (ECG) is the best popular and low-cost investigative device for assessing the heart’s electrical impulses and diagnosing cardiovascular health. The heart beats in a regular rhythm, with regular myocardial excitation, circulating blood to the entire body. During the practice, When the myocardium contracts, a little current is generated by the heart and carried to the surface of the body, generating potential modifications in every portion of the human body. The ECG is created by calculating the potential drift with the help of electrodes placed on the limbs and chest of the subject and recorded with electrocardiograph or a vector electrocardiograph (Liu et al. 2021). The aberrant activity of the heartbeat and rhythm can be demonstrated in this way. CVD, Coronary heart disease, and congestive heart failure can all be predicted using an ECG. Because several of these disorders are linked to an elevated liability of stroke or sometimes it can lead to death, early detection is critical. ECG has been shown to be helpful in determining both short- and long-term results in investigations. For individuals with MI, for instance, the sooner the irregular cardiac rhythm is diagnosed, the better the possibility of avoiding life-threatening complications and recovery (Lown et al. 1969). Hence the agile and precise diagnosis of ECG is essential clinically. Since signal analysis is a time-consuming and difficult operation, there is a risk of personal ambiguity and human mistakes during the analysis procedures, also for professionals who have been taught for years. As a result, it is critical to experiment with computer-assisted methods. Computer-assisted exploration can evaluate ECG signals more accurately and quickly without causing any differences (Liu et al. 2021).
The first computer-assisted ECG analysis, structure was created in the 1960s (Pipberger et al. 1961). Data preprocessing, feature extraction (FE), and categorization are the three essential processes in the completely automatic structure of the classic intelligent procedure for the analysis of ECG. In the preprocessing stage, data is denoised, added, or chopped into signal slices of the same length. For ECG signal categorization, the FE phase is critical. The geometry of the ECG signal not only in the time domain but also in the frequency domain, as well as the cardiac rhythm, can be utilized to excerpt features. Finally, based on the collected attributes, the signals are categorized as distinct types of heartbeat or sickness. The vision is to build algorithms that are exceptional in accuracy, efficiency, and stability while also reducing the strain on doctors.
Deep learning (DL), is a computer-assisted process with a significant capability of extracting features, was able to classify ECG signals with high accuracy (Murat et al. 2020). DL is accomplished through the formation of hierarchical artificial neural networks (Goodfellow et al. 2016). Deep learning, according to (Bengio et al. 2007), discovers subtle structure in big datasets by utilizing the backpropagation technique to determine how a model's internal weight values, which are utilized to generate the illustration in every layer, should evolve. As a result, DLs are allowed to possess fine culpability tolerance and avoid overfitting-induced errors. DL can automate FE and categorization by emulating the overall purpose of studying the human brain, whereas, in the past, these tasks required engineers to plan. In this approach, it can be learned that implicit knowledge was previously mastered by professionals, implying a significant human burden. The advancements in the central processing unit (CPU) reduce the execution and training time effectively. As a result, DL can now train enormous volumes of data and use more complicated algorithms, providing it with much more development potential.