Research On Fetal Electrocardiogram Extraction From Abdominal Electrical Signal Of A Pregnant Woman

In the world,there is no clinical fetal electrocardiograph which has been accepted by hospitals.The prime reason is that the electrical signal acquired from the abdomen of a pregnant woman is a nonstationary and nonlinear complicated signal,with very low signal noise ratio(SNR),and it is very difficult to extract the fetal ECG(FECG)from the abdominal ECG signal.In the acquired abdominal signal,the fetal ECG component is much smaller than the maternal ECG(MECG)component,and it is also mixed with the 50Hz power line interference,maternal electromyogram(MEMG),fetal electromyogram(FEMG)and other random electronic noises.The FECG component is seriously contaminated,even masked by the interference and noises.Approximately 10%?30%of the maternal QRS complexes and the fetal QRS complexes coincide in time waveform,and most of the maternal harmonics and fetal harmonics superpose in frequency domain.The contingent and discrepant mEMG or fEMG makes the abdominal ECG signal have complicated nonstationarity.The complex propagation paths of the mECG and FECG from the hearts to the body surface bring about abdominal ECG signal strong nonlinearity.How to extract the FECG from the maternal abdominal signal effectively is always a hot challenging problem at home and abroad.The study on this issue is carried out intensively in this dissertation.The research content and points of innovation are shown as follows:(1)An adaptive modelling in phase space for fetal ECG extraction is proposed.The phase space trajectory is reconstructed from one abdominal ECG signal and multi-leads thoracic ECG signals.The nonlinear relationship between the thoracic ECG signals and maternal ECG component in the abdominal signal is built by the local adaptive modelling,and the maternal ECG component in abdominal ECG signal can be estimated,which is then subtracted from the abdominal ECG signal to obtain a fetal ECG estimate.The adaptive modelling process carries on the spatial neighboring points of each point in the phase space,not on the temporal neighboring points of each value in time sequence as usual.This can utilize the geometric information of the ECG signals,and the simpler local relation represented in phase space can make the estimation of the nonlinear model easier and more accurate.Compared to two other multi-lead methods,classical adaptive filtering and ICA,the proposed method has shown improved robustness and fidelity in restoration of the FECG during testing with synthetic ECG signals and a real fetal ECG database.(2)An improved local principal component analysis(LPCA)in pseudo phase space for fetal ECG estimation from a single abdominal ECG signal is proposed.The parameters of classical LPCA method are determined by artificial visual inspection of the ECG waveform,and it is inconvenient,less robust and inefficient.These flaws have been overcomed in this dissertation.We divide the pseudo phase space into two regions:original point region and QRS wave region.The processing parameters are different for two regions and they are determined automatically according to the ECG waveform.Most of the points in pseudo phase space belong to the original point region,and we use batch processing mode to clean these points.This procedure decreases the computer-time intensively compared with the classical approach.The improved LPCA method has shown the better robustness and efficiency when testing with synthetic ECG signals and a real fetal ECG database.It is also found that the low-frequency band of the fetal ECG would be lost using LPCA method,and the positions of the extracted fetal R peaks are accurate,thus it can be used for the fetal heart rate(FHR)estimation.(3)A combinative method including R-peak detection,resampling and comb filter has been proposed to estimate the FECG from a single-lead abdominal signal.ECG signals are nearly periodic with repetitive patterns,and the durations of the R-R intervals vary with time due to heart rate variability(HRV).It is consequently not possible to separate the ECG components using a comb filter directly.We can make the sample length of the each R-R interval uniform by changing its sampling frequency,which can be accomplished by R-peak detection and resampling process.After that,a comb filter can be applied to the resampled signal to extract the ECG component,and all of the R-R intervals of the filtered waveform can be resampled again to recover their original sampling frequency.The proposed method can extract the maternal ECG component and fetal ECG component subsequently from the abdominal ECG signal.Compared to two other single-lead-based methods,the proposed method has shown improved robustness and fidelity in restoration of the FECG during testing with synthetic ECG signals and a real fetal ECG database.(4)A template based adaptive modelling in phase space for single-lead fetal ECG extraction is proposed.The template signal of the objective ECG component can be obtained from a single-lead ECG signal.The original ECG signal and template signal are reconstructed in a phase space,and the objective ECG component is estimated by the local adaptive modelling in phase space.Compared to the multi-lead adaptive modelling in phase space proposed in section 2,this method has three merits:(1)only one abdominal ECG is needed;(2)the nonlinear relationship to be fitted is simpler in local phase space;(3)the further extraction of fetal ECG is available to obtain a clearer fetal ECG signal.It has been proved that the proposed method can obtain the reasonable estimation of fetal ECG even in the low SNR condition and testing with real database.All of the proposed four methods have been tested with the Non-Invasive Fetal Electrocardiogram Database from the MIT-BIH PhysioBank.The real FECG signals are highly contaminated by the mECG and noises with very low SNR,and the extraction of the FECG is very hard.As far as we know,there was no literature to use the real database to test their algorithm overall before the authors' work.The comb filter based method and template based adaptive modelling in phase space method show the top performance with the nearly the same success rates 78%,which seems to be a promising result.The improved LPCA in pseudo phase space method and the multi-lead adaptive modelling in phase space method have relatively worse performance with the success rates 51%and 49%respectively,but cost less computation time.