A Novel Approach to Cardiac Arrhythmia Detection Using ECG Recurrent Plots and Convolutional Neural Networks

Cardiovascular disease is the most common disease worldwide because of which several people die annually. Recognizing cardiac arrhythmia is of great importance as several lives will be saved. Recent studies have tried to recognize arrhythmia and classify electrocardiogram (ECG) signals using traditional classifiers. Deep Learning (DL), which is considered a milestone in machine learning, has recently attracted attention in different fields including healthcare. Although it needs huge data and is time-consuming, higher performances achieved by DL models have motivated researchers to use DL in their studies. ECG as a reflection of a complex nonlinear system (i.e. human heart) is complex and nonlinear. We aim to employ ECG nonlinear analysis to extract meaningful information and achieve higher classification rates. In this paper, we propose a novel method using ECG recurrence quantification analysis and convolutional neural networks. ECG signals, from a reliable dataset, are employed to reconstruct recurrence plots. Convolutional neural networks (CNNs) are utilized to classify ECGs into two groups including "normal" and "arrhythmia". Results suggest that our proposed method is effective in comparison with previous studies and will help researchers with their future studies in the field of cardiac arrhythmia detection.

Keywords:

Cardiac Arrhythmia Detection , Deep Learning , Electrocardiogram , Recurrence Quantification Analysis , ECG Nonlinear Processing.