A Deep Neural Network Approach to Automatic Multi-Class Classification of Electrocardiogram Signals

Abstract

Classification of electrocardiogram (ECG) signals is essential for accurate clinical diagnosis of coronary illness. Deep Neural Network (DNN) has emerged as a promising tool for feature identification in ECG signals with advancements in signal processing techniques and artificial intelligence. Particularly, Convolutional Neural Network (CNN) has enabled intricate feature identification and classification of images. In this scope, an efficient classification methodology utilizing wavelet transform and AlexNet CNN is presented in this paper to classify the ECG signals into three different categories, namely: Arrythmia (ARR), Congestive Heart failure (CHF) and Normal Sinus Rhythm (NSR). Physionet research dataset is utilized in this work. Firstly, all ECG signals are converted to a corresponding scalogram using Continuous Wavelet Transform (CWT). Then, these scalograms are applied as input to AlexNet, which is a pretrained CNN. Finally, transfer learning and fine-tuning are done, and the performance of modified AlexNet is evaluated. The study revealed a good accuracy of 97.3%.