Noise-Tolerant Neural Network Approach for Electrocardiogram Signal Classification

Abstract

Several machine learning approaches have been proposed for classifying electrocardiogram (ECG) signals. Most of these use adaptive filtering techniques to reduce the noise corruption embedded in the signals. However, band-pass filters can affect the estimation of morphological parameters and result in misleading interpretation. We propose a noise-tolerant neural network (NN) approach, based on artificial noise injection, to improve the generalization capability of the resulting model. The NN classifier initially discriminated normal and abnormal heartbeat patterns in simulated ECG signals with minor, moderate, severe, and extreme noise, with an average accuracy of 99.2%, 95.1%, 91.4%, and 85.2% respectively. Ultimately we performed a fairly accurate recognition of four types of cardiac anomalies for single lead of raw ECG signals, obtained an overall classification accuracy of 95.7%. Therefore, is a useful tool for the detection and diagnosis of cardiac abnormalities.