Fetal ECG Signal Extraction Based On FastICA Blind Separation Theory

Fetal Electrocardiogram(FECG)refers to the electrical signal of the fetal heart that has periodic changes due to cardiac contraction and relaxation during the activity.It can intuitively reflect the health status of the fetus during pregnancy.It is crucial to conduct medical judgments.The difficulty in FECG extraction is that the abdominal signal collected by the non-invasive method is not pure.The abdominal signal contains a very weak FECG component which is completely submerged in the maternal signal and various noises.And there is no any priori knowledge about FECG and MECG(Maternal electrocardiogram).The principle and research status of various FECG extraction algorithms is summarized firstly in this thesis,and each wave group in FECG waveform and two acquisition methods of FECG are introducesd.The research in this thesis is aimed at the mixed signal of the abdominal wall collected by non-invasive method,and the characteristics of the signal are introduced.In this thesis,the following research work has been done on how to extract clear FECG:1.Firstly,the basic theory related to ICA is discussed.The basic model of ICA is analyzed in this thesis,and the process of preprocessing the ICA algorithm data is given: de-average,whitening and dimensionality reduction.Three kinds of objective functions have been introduced which is commonly used in ICA algorithms.It focuses on the principle of using non-Gaussian as the objective function to measure the independence of separated signals.Two problems exist in the ICA algorithm are analyzed: the separated signal amplitude is undeterminable and the signal sequence extracted by the algorithm is undeterminable.2.The FastICA algorithm based on negative entropy is introduced to extract FECG.The definition of negative entropy is given in this thesis.In order to better apply negative entropy to practical applications,the negative entropy is approximated and the FastICA algorithm based on negative entropy is constructed.For the basic algorithm,it is sensitive to the initial value.The truncation method and iterative formula of the second-order convergence are used to replace the derivation in the calculation,which constitute two improved algorithms.An iterative formula of fifteen-order convergence that is insensitive to the initial value is introduced in the thesis,and then obtains a third improved algorithm.Four kinds of algorithms are tested by using synthetic ECG data and clinical ECG data.The performance of the four algorithms is analyzed from iteration number,iteration time,signal-to-noise ratio and PI index.3.For the problem that the extraction order cannot be determined in the ICA algorithm,the sample entropy is used to solve this problem.The sample entropy reflects the probability of new information appearing in the signal.The smaller the value of the sample entropy,the signal is more regular;the larger the value of the sample entropy,the signal is the irregular,and the probability of generating new information is greater.The simulation results of MATLAB platform verify that the sample entropy value corresponding to FECG is the largest.The sample entropy can be calculated by separating the separated independent components to find the channel where FECG is located.4.The basic algorithm and the third improved algorithm are used to test the simulated ECG signal and the clinical ECG data,and the separated FECG is subjected to wavelet denoising processing.The signal-to-noise ratio of FECG separated before and after noise reduction from the synthesized ECG data as the source signal proves that wavelet denoising can improve the signal-to-noise ratio of FECG,resulting in a purer FECG.