Image Based ECG Signal Classification Using Convolutional Neural Network

Abstract

Electrocardiogram (ECG) analysis is one of the gold standards in diagnosing heart abnormalities. Commonly, clinicians analyze the ECG signal visually by observing the shape, rhythm, and voltage of the signal. Some of them are supported by the application of automatic diagnosis of the ECG device itself. Currently, digital signal processing combined with traditional or advanced machine learning plays an important role in supporting medical diagnosis including ECG diagnosis. However, it is often constrained by the lack of raw data support from most commercial ECG devices. Classification method by processing ECG image can be one way to tackle this problem. Therefore, in this preliminary study, an image-based ECG classification method using a deep learning approach is proposed. The ECG signals analyzed in this study include normal sinus rhythm (NSR), premature ventricular contraction (PVC), and Bigeminy. Convolutional neural network (CNN) with VGG16 architecture has been employed for feature extraction and classification. The simulation results show up to 95% accuracy in detecting ECG abnormalities. The results of this study can be an alternative in detecting ECG abnormalities and can be considered as a supporting diagnosis by the clinician.