Research On Fetal Electrocardiogram Extraction With Blind Source Separation Method

The fetal electrocardiogram (FECG) signal, a graphical recording of the electrical potentials generated in association with fetal heart activity, is one of many important physiological signals commonly used in clinical practice. As in adults, the well-being and status of a fetus can be assessed from a fetal ECG signal. Comparing with the pononcardiogram and heart beat signals, fetal electrocardiogram better reflects the complete view of heart activities. And it is more sensitive than Doppler ultrasound Fetal Heart Rate (FHR) in case of fetal acidosis and anoxia. However, the FECG signals acquired from the electrodes attached on the maternal abdomen usually contaminated by strong maternal electrocardiogram (MECG) signals, which will bring difficulties to medical diagnosis. Extracting FECG from the mixed ECG abdominal signals is the primary objective in biomedical signal processing area.Accroding to the blind source separation (BSS) techniques, several new FECG extraction methods are proposed.A FECG blind source extraction method base on kurtosis and skewness is introduced. The high order statistical characteristics– kurtosis and skewness of the MECG and FECG are estimated from cilinic signals. The results show that the kurtosis of both is comparable, but the absolute skewness is quite different. Adding a skewness item to the objective function of FastICA, a new objective function combined kurtosis and skewness with a proportional coefficient is created. Theoretical analyses prove the local convergence and local stabilization of the algorithm. The range of the proportional coefficient combined the kurtosis and skewness of the signals is determined by experiments. The simulations with simulated and real clinical signals show that the proposed algorithm is better than FastICA, but slightly worse than JADE.As the linear mixing model of multi-channel BSS is difficult to satisfy in practice, an oblique projector based FECG blind source extraction method with single observation is described. ECG signal is a type of quasi-periodical signal. It implies the strong correlation between adjacent heart beats, which means that there are only tiny difference appeared in a relatively short period.Therefore, a new delay-accumulation model of ECG is proposed to apply to the oblique projection. In order to satisfy the condition of oblique projector, MECG and FECG signals from normal pregnancies are aquired to study the rank of the shift matrix and the matrix pencil of the signals. Simulation with clinical signals shows that the oblique projector based single channel FECG blind source extraction method can very effectively separate FECG signals from MECG signals.Because the oblique projector based single channel FECG blind source extraction method needs the precise location of each FECG complex, a method to localize FECG comlex from single channel observation, aided with reference signal, such as phonocardiogram is proposed. It is diffcult to identify the FECG comlex overlapped by a MECG QRS comlex. Errors will be introduced to the MECG comlex, when the overlapped FECG is mis-identified. Thus, minimizing the errors in MECG will guide to the true position of the FECG comlex. Instead of directly detecting weak and low-SNR FECG used by other FECG identifying methods, the proposed method inspects and tests the strong and high-SNR MECG signals to identify the FECG comlex, which improve the reliability and accuracy of the FECG detection. Experiments with simulated and real clinical signals show that the method can localize the overlapped FECG accurately.