Fetal Electrocardiogram Extraction Based On Semi-Blind Deconvolution Source Separation

The fetal electrocardiogram(FECG) is one of the most effective methods of perinatal fetal monitoring. It could help doctor to discover the fetus disease during intrauterine development by the analysis of FECG waveform, thus reducing the injury rate and mortality of newborn babies. However, the signal-to-noise ratio of FECG in abdominal electrocardiogram(ECG) signal observed by noninvasive method is low and the FECG signal is often contaminated by maternal ECG(MECG) and other strong noise interference, so how to extract FECG signal clearly is always an important research subject of the fetal monitoring. Blind source separation(BSS) has been regarded as having the best perspective among all FECG extraction algorithms, existing FECG extraction algorithm based on BSS usually constructs linear instantaneous model without any ECG feature information, which causes the extracted result is low accuracy and has obscure physiological significance. In this paper, considering the aspect of signal characteristics and model, the semi-blind deconvolution(S-BD) source separation method based on convolution mixture was presented to extract clear FECG signal as complete as possible. The specific content is studied as follows:(1) Analyzing FECG characteristics and constructing linear convolution mixturemodel.Analyzing FECG characteristics by starting with the electrophysiological characteristics, the rationality of blind deconvolution method for extracting FECG was confirmed by the non-minimum phase property of FECG signal; Combined with the actual ECG characteristics and blind deconvolution source separation mixture model, abdominal ECG signal was constructed as linear convolution mixture model and FECG extraction method based on ECG signal feature characteristics was proposed.(2) Based on cyclostationary characteristic and convolution mixture model, the S-BD source separation algorithm was presented.S-BD was analyzed in detail to extract the FECG signal. The time delay and cyclostationary parameters were adopted in the separation target function, which be used to measure the non-Gaussian maximization. The target function was optimized by the gradient method to realize the MECG and FECG monophyletic signal separation. To improve the FECG estimation accuracy, the MECG contribution elimination algorithm based on the least square inverse filtering was used to improve the time-domain subtraction of traditional BSS. The experiment result shows that MECG contribution elimination algorithm keeps the FECG waveform more complete than time-domain subtraction for the temporal overlap of maternal and fetal complexes.(3) Based on the S-BD source separation algorithm, a FECG extraction complete method was proposed.Combined with single-source separation and MECG contribution elimination algorithm, the FECG extraction complete method including preprocessing, parameter estimation, feature extraction and enhancement was proposed. Analyzing the FECG extracted by the S-BD source separation and the traditional BSS method. Experimental simulation results show that the fetal heart rate and waveform of FECG extracted by S-BD source separation method was extremely close to the reference FECG acquisited by fetal scalp electrodes; and S-BD source separation performance was better than the traditional blind source separation in visual waveform and quantitative performance like signal noise ratio(SNR) and so on.