Diagnosing Heart Disease Using Deep Learning Techniques

Abstract

The Electrocardiogram (ECG) is one of the most important body bio-signals, through which many heart and blood circulations disorders and abnormalities can be detected. So that, many studies in this field have been done like making systems that read and analyze ECG signals using artificial intelligence algorithms. This project proposes a Convolutional neural network that we build from scratch. This CNN read and process ECG signals that we get from the MIT-BHA dataset. This dataset contains about 87554 ECG samples.

Samples are sorted in 5 classes, Normal beats, Supraventricular ectopic beats, Ventricular ectopic beats, Fusion beats and Unknown beats. Signals were processed and used in our CNN training and testing procedure. The confusion matrix derived from the testing dataset indicated 97% accuracy for 'normal' class.  For the “Supraventricular ectopic beats” class, ECG segments were correctly classified 85% of the time. For the “Ventricular ectopic beats” class, ECG segments were correctly classified 95% of the time. For the “Fusion beats” class, ECG segments were correctly classified 88% of the time. Finally for the “Unknown Beats” class, ECG segments were correctly classified 98% of the time. In total, there was an average classification accuracy of 96.64%.