Arrhythmia Detection from Heart Electrical Signal Using 1-D Convolution Neural Network

The World Health Organization (WHO) claims that heart conditions have a significant negative impact on public health. The statistics show that 17.9 million people lose their lives each year to heart conditions. Medical institutions employ electrical heart impulses to diagnose arrhythmia. The electrocardiogram (ECG) is the most often employed diagnostic tool by medical practitioners. Cardiologists analyze the recorded ECG signal visually, this process requires time and is dependent on the doctors’ level of expertise. A computational method for very accurate automatic detection may be useful for clinical specialists. A machine learning approach can automatically classify arrhythmias, and feature extraction enhances classification performance. Deep learning systems can automatically extract features. A convolutional neural network is employed for the automatic classification of arrhythmia from the heartbeat into five types. Unknown beats (Q), fusion beats (F), supraventricular ecotic beats (S), ventricular ecotic beats (V), and non-ecotic beats (N). For research purposes, data is collected from the MIT-BIH arrhythmia database. The proposed attention-based Convolutional Neural Network (CNN) model has a classification accuracy of 99.61%.