A Cubature Kalman Smoother for ECG Denoising

Abstract

Model-Based Bayesian frameworks have been used extensively in the field of ECG processing. In this paper, an ECG denoising framework based on cubature Kalman smoother (CKS) is proposed. In addition, we used dynamic time warping (DTW) in the ‘ECG phase-wrapping’ stage of this framework to improve its performance in case of heart rate variability. The proposed filter was evaluated on several normal ECG segments extracted from MIT -BIH normal sinus rhythm database (NSRDB). To do so, artificial white Gaussian and non-stationary real muscle artifact (MA) noise over a range of SNRs from 10 to −5 dB were added to these normal ECG segments. The benchmark methods were the extended Kalman filter (EKF), extended Kalman smoother (EKS), unscented Kalman filter (UKF), Unscented Kalman smoother (UKS) and cubature Kalman filter (CKF) frameworks. Among the benchmark algorithms the EKF framework is the first and CKF is the most recent model-based Bayesian algorithms proposed for ECG denoising. The results showed that the proposed algorithm had a noticeable advantage over EKF, UKF, and CKF methods from SNR improvement viewpoints at all input SNRs. The results also revealed that UKS and CKS perform similar to each other and at low input SNRs, these two algorithms outperform the EKS algorithm.