A Novel Approach for Arrhythmia Classification Using CI- 1D-LBP with LSTM, 1D-CNN and GRU Models

Abstract

Atrial arrhythmias (ARR) are known as the most encountered cardiac disorders in today's world that have direct or indirect detrimental effect on human health. Therefore, Computer-Assisted Diagnosis (CAD) systems are instrumental in the early detection and diagnosis of diseases, serving a pivotal role in the initial assessment and identification process. In this study, ECG data belonging to four different types of arrhythmias, namely ventricular beat (VB), supraventricular beat (SVB), fusion beat (FB), and an unidentified arrhythmic beat (UB), as well as ECG data showing normal sinus rhythm (NSR) of healthy individuals were classified. The ECG data were sourced from the MIT-BIH database. The Center-Independent 1Dimensional Local Binary Pattern (CI-1D-LBP), originated from the local binary pattern (LBP) method, proposed as a new approach for deriving the essential features needed for the classification of ECG signals. With this new approach, histograms are generated from the signals, and these histogram data are used as input for classification in 1D-CNN, LSTM, and GRU deep learning methods. The CI-1DLBP+GRU methodology exhibited superior efficacy in classifying the five-labeled dataset (VB-SVB-FB-UB-NSR) relative to the other applied methods, attaining an impressive accuracy rate of 98.59%.