Cancellation of Artifacts in ECG Signals Using a Normalized Adaptive Neural Filter

Abstract

Denoising electrocardiographic (ECG) signals is an essential procedure prior to their analysis. In this paper, we present a normalized adaptive neural filter (NANF) for cancellation of artifacts in ECG signals. The normalized filter coefficients are updated by the steepest-descent algorithm; the adaptation process is designed to minimize the difference between second-order estimated output values and the desired artifact-free ECG signals. Empirical results with benchmark data show that the adaptive artifact canceller that includes the NANF can effectively remove muscle-contraction artifacts and high-frequency noise in ambulatory ECG recordings, leading to a high signal-to-noise ratio. Moreover, the performance of the NANF in terms of the root-mean-squared error, normalized correlation coefficient, and filtered artifact entropy is significantly better than that of the popular least-mean-square (LMS) filter.