A survey of deep learning approaches for classifying ECG heartbeat arrhythmias

An automated computer aid remains relevant to support cardiology specialists in diagnosing heart disorders and rapidly classifying arrhythmias by using an electrocardiogram (ECG), which is among the most regularly utilized techniques to identify health disorders because hand identification of these heart-beat classes by doctors might take a long time. In this paper, we investigated and reviewed diverse research that worked on ECG arrhythmia identification by employing deep learning approaches. We illustrated and discussed the performance and approaches adopted to identify ECG heart arrhythmias by six commonly utilized methods, including MLP (Multilayer Perceptron), CNN (Convolutional Neural Network), DBN (Deep Belief Network), RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), and GRU (Gated Recurrent Unit). We considered various limits and disclosed that there is yet space to extend the classification’s performance, precisely by reducing the preprocessing and principally the computational expense. Such a managed paper survey provides specialists with a nearly clear view of some elements of ECG classification methods and allows them to explore points unmet until now.

I. Introduction

A vital organ in the human body, which is the heart, pumps blood to all of the body’s organs, including the heart, providing oxygen and nutrition while also eliminating useless waste. It is susceptible to illness and diseases just like every other organ in the body. Cardiac rhythm abnormalities are among the most common signs of heart disease. It manifests itself when the heart rate is abnormally fast or slow, or when the heart rate is lower than normal. Ectopic rhythms can be caused by irregularities in the heart’s sinoatrial (SA) node or inconsistent pacing action from different areas of the heart. The majority of arrhythmias are innocuous, but they can appear as a result of a sudden and potentially fatal disease. Arrhythmias are divided into two types: those that require long-term treatment (such as myocardial infarction, tachycardia, ventricular fibrillation, etc.) and those that require prompt treatment as a preventive strategy. Cardiovascular diseases (CVDs) are the long-term consequences of cardiac arrhythmias, and as a result, they pose an immediate threat to a person’s life (e.g., ventricular fibrillation and tachycardia), which can end in death. In 2016, cardiovascular disease was the leading cause of death in humans, representing 31% of mortality worldwide [1], with heart attacks accounting for 85% of all deaths. The situation will become increasingly visible in regions where medical professionals and clinical materials are in limited supply. This increases the necessity for a reliable, automated, and inexpensive control and diagnosis technique. Healthcare providers are increasingly requesting this need so that they may integrate medical evaluations and employ computer-aided diagnosis systems (CADS). A computer-aided diagnosis system is a set of automated health monitoring systems that work by analyzing physiological signals to control and evaluate the functionality of the relevant organ. CDSs enable patients with simple and practical ways to learn more about their illnesses, which can help medical professionals treat them more effectively.