Cardiac Arrhythmia Diagnosis Using Deep Learning: A 1D CNN-GRU Approach with Multiclass SVM and DWT Analysis

Arrhythmia is a cardiovascular disease characterized by an abnormal heartbeat in rate or rhythm. An electrocardiogram (ECG) is an important part of cardiac arrhythmia diagnosis in clinical practice. Recording ECG for the detection of arrhythmias can pose challenges under certain circumstances. This paper's objective is to employ deep learning methods in cardiac arrhythmia diagnosis based on ECG signals, with the least amount of data pre-processing. The proposed algorithm architecture is characterized by pre-processing, feature extraction and 1-Dimension Convolutional Neural Network (1D CNN) Gate Recurrent Unit (GRU) classification for arrhythmia detection. In this paper, the following arrhythmia categories have been considered: Normal beats, premature contractions originating in the ventricles (PVCs), disruptions in the electrical conduction through the left bundle branch (LBBB), interruptions in the conduction through the right bundle branch (RBBB), and artificially initiated depolarizations (paced beats). Classifiers produce vectors, and the highest element in each vector indicates the degree of membership in the corresponding class. The proposed approach involves employing a 1D CNN coupled with GRU for classification, alongside a multi-Class SVM, resulting in an impressive accuracy of 99.97%. Though the characteristics of Discrete Wavelet Transform (DWT) are similar, their output from Multiclass Support Vector Machine (Multi SVM) is hardly different.