A novel weighted approach for automated cardiac arrhythmia beat classification using convolutional neural networks

Abstract

Arrhythmia is a cardiac disorder in which the normal blood pumping activity of the heart becomes irregular. This heart malfunction can result in serious heart disease and even death. Therefore, detection and proper treatment of arrhythmia are essential. The abnormal heart behaviour can be recorded using an electrocardiogram (ECG). A one-dimensional convolutional neural network (1D-CNN) with a novel weighted approach was proposed to detect and classify arrhythmia types from ECG signals. The proposed classifier was trained and evaluated using the Massachusetts institute of technology-Beth Israel hospital (MITBIH) arrhythmia database to classify five arrhythmia beat categories (N, S, V, F, and Q), as recommended by the Association for Advancement of Medical Instrumentation (AAMI). The proposed model obtained an overall sensitivity of 94.35%, precision of 94.02%, specificity of 99.5%, and accuracy of 99.65%. The experimental results demonstrate that the proposed CNN model can achieve cutting-edge performance and can be used for arrhythmia diagnosis in real-time.