Automatic electrocardiogram sensing classifier based on improved backpropagation neural network

Abstract

As heart disease is among the common diseases endangering human life, the electrocardiogram (ECG) recognition of various categories of abnormal heartbeat rhythms is essential for boosting the success rate of treatments for this illness. In this paper, we propose an automated ECG recognition method based on a backpropagation (BP) neural network. First, biorthogonal (bior) wavelet denoising was adopted to eliminate baseline drift as well as highfrequency noise in the ECG. Then, a dyadic spline wavelet was used to detect the QRS, T, and P waves. Six amplitude features and 15 range features were extracted to better represent the local and global features of the ECG, respectively. Finally, the optimum BP neural network (BPNN) model was utilized to identify the ECG signal. The optimum BPNN model exhibited a steady precision of more than 99% in the recognition of ECG signals, which is superior to the results for a support vector machine (SVM) and a convolutional neural network (CNN), with a significantly improved correct recognition rate of type 5 ECG signals.