As cardiovascular diseases (CVDs) are a serious concern to modern medical science to diagnose at an early stage, it is vital to build a classification model that can effectively reduce mortality rates by treating millions of people in a timely manner. An electrocardiogram (ECG) is a specialized instrument that measures the heart's physiological responses. To accurately diagnose a patient's acute and chronic heart problems, an in-depth examination of these ECG signals is essential. The proposed model consists of a convolutional neural network having three convolutional, two pooling, and two dense layers. The proposed model is trained and evaluated on the MIT-BIH arrhythmia and PTB diagnostic datasets. The classification accuracy is 99.16%, which is higher than state-of-the-art studies on similar arrhythmias. Recall, precision, and F1 score of the proposed model are 96.53%, 95.15%, and 99.17%, respectively. The proposed model can aid doctors explicitly for the detection and classification of arrhythmias.
TopBackground
This section describes, heart, ECG signal, various cardiovascular abnormalities, and a comparison of various approaches used for the classification of arrhythmias.
Heart, a muscular organ, is responsible for the supply of the blood to all other human body parts (Aje, 2009). As seen in Fig. 1, a blockage in a coronary artery, which feeds blood and oxygen to the heart, causes a heart attack, which is a major health risk.
Figure 1. Heart attack or Myocardial ischemia
Figure 2. The standard Electrocardiogram (ECG) signal.
The characteristics of a normal ECG signal are depicted in Figure 2. Where the P-wave and QRS-complex represent depolarization/contraction of the atrium and ventricles, respectively, while the T-wave represents ventricle repolarization. The ECG signal is derived from an electrocardiogram (ECG), a device that records the heart’s physiological activities for a time interval. ECG data is used to diagnose various cardiovascular abnormalities, such as premature atrial contractions (PAC), myocardial infarction (MI), atrial fibrillation (AF), premature ventricular contractions (PVC), congestive heart failure (CHF) and supraventricular tachycardias (SVT). A rapid development of portable devices such as Holter monitor (Nikolic et al., 1982), Apple watch etc., are developed to detect the ECG signal in recent years, which resulted in a big datasets for the researchers and human cardiologists to analyse the ECG data more accurately. As a consequence, automatically and reliably analysing ECG data has become a popular research topic. ECG data can also be used in innovative ways such as sleep staging and biometric human identification. To manage the large datasets, we need an efficient approach for the analysis. As a result, the datasets are subjected to machine and deep learning algorithms (Hiran, K. K.et al.2021). Figures 3 and 4 shows the basics of machine and deep learning approaches respectively (Xia et al., 2018). Traditional machine learning approach requires experts (human cardiologists) to extract characteristics from raw data and then generate final outputs using machine learning models or decision rules, whereas deep learning uses deep neural networks and extracts features automatically (Mahrishi, M.,et al.,2020). Here, expert features are a) Statistical features (heart rate variability, density histograms, sample entropy etc.), b) Frequency-domain features (Romero et al., 2001), c) Medical features (P, Q, R, S, and T wave parameters).
Figure 3. Traditional machine learning.