Noise removal from electrocardiogram signal employing an artificial neural network in wavelet domain

Electrocardiogram (ECG) signal involves significant information about heart state and is one of the common tools for cardiologist in diagnosis of heart failures. Using adaptive filters for filtering this signal, which inherently has nonstationary features, is used as one of the known methods. In this paper, the wavelet transform and also a neural network (NN) based on adaptive filters are used for removal of undesirable noise from the ECG signal. In this context, in training stage, network weights related to each wavelet sub band is obtained by using the steepest descent algorithm, and filter coefficients for removal of noise from ECG signal are calculated. Results obtained from employing this algorithm on the MIT-BIH database and simulated ECG signal are indicative of improved performance of noise removal in comparison with other methods.