Single-channel fetal ECG extraction method based on extended Kalman filtering with singular value decomposition algorithm

In this paper, an approach based on extended Kalman filtering (EKF) combined with singular value decomposition (SVD) was proposed to extract fetal electrocardiogram (fECG) from single-channel abdominal ECG recording. First, on the basis of dynamic model of maternal electrocardiogram established, the maternal ECG component was estimated from the single-channel abdominal signal by EKF or extended Kalman smoothing (EKS). Next, the preliminary estimation of fetal ECG was obtained by subtracting the maternal ECG component from the single-channel abdominal signal. To obtain the fetal ECG with high signal-to-noise ratio, the preliminary fetal ECG was denoised using SVD. In addition, an improved construction method of the ECG signal construction matrix was proposed in the SVD algorithm for the case of fetal arrhythmia. Synthetic abdominal signals and real abdominal signals (49 abdominal channels in total from DAISY database and PhysioNet noninvasive fetal ECG database) were used for fetal ECG extraction. The results showed that the signal-to-noise ratio of fetal ECG extracted by EKF + SVD or EKS + SVD was about 5 dB higher than that by EKF or EKS alone, and the accuracy of fetal ECG extraction was 95.60% and 95.94%, respectively. In conclusion, the signal-channel fetal ECG extraction method based on the combination of EKF and SVD could effectively improve the signal-to-noise ratio and accuracy of fetal ECG signals extraction, and would be suitable for maternal or fetal arrhythmia.