A robust iterative nonlocal means method for electrocardiogram signal denoising

As a noninvasive technique, electrocardiogram (ECG) plays a significant role in the diagnosis of many cardiac diseases. However, ECG signals are usually corrupted with baseline wander and high-frequency noise. Numerous approaches have been proposed to suppress noise in ECG signals such as finite impulse response (FIR) filter, infinite impulse response (IIR) filter and wavelet method. These methods cannot provide sufficient noise suppression or preserve details very well. The recently proposed nonlocal means (NLM) method can overcome this drawback to some extent, but it cannot denoise the ECG signals effectively at high noise corruption due to the adoption of the noisy signals for determining the weight. To address this problem, a robust iterative nonlocal means (INLM) is proposed in this paper. The proposed method uses an iterative strategy to denoise ECG signals, in which the weight is determined by replacing the noisy signals with their denoised versions. Extensive experiments on the simulated and clinical ECG signals demonstrate that the proposed method can remove noise effectively while preserving useful details of ECG signals very well, and it outperforms above-mentioned denoising algorithms in terms of objective metrics such as root-mean-squared-error (RMSE) and signal-to-noise ratio (SNR).