Study On Fetal ECG Extraction Based On ICA And NMF

Standardized and effective prenatal fetal monitoring is a real-time dynamic observation of fetal growth and health,and the commonly used monitoring means is fetal ECG monitoring.Fetal ECG signal is the source of biological electrical signal during fetal heart activity,which contains abundant information.Because the fetal electrocardiogram(FECG)signals collected from the mother’s abdomen are often accompanied by interference noises such as mother’s ECG signal,power frequency interference,baseline drift and mother’s EMG,how to extract accurate FECG signals has been a major difficulty for medical researchers.FastICA based on negative entropy combines the statistical characteristics brought by negative entropy and the excellent algorithm characteristics of fixed point iteration.It is a fast optimization and iteration multi-dimensional data processing technology.ECG signals collected from the body surface of pregnant women can be regarded as a mixture of multiple independent signals,which meets the precondition of ICA algorithm.So FastICA algorithm can be used.Separate each source signal.Non-negative matrix factorization(NMD)is an unsupervised and effective dimensionality reduction method for processing data,which can further extract noisy fetal ECG.Therefore,in view of the excellent theoretical results of predecessors,combined with simulation experiments,a fetal ECG signal extraction method based on independent component analysis and non-negative matrix decomposition is proposed.Three kinds of multi-channel ECG data were used in this experiment,including simulated ECG signal,international general MIT-BIH ECG database and clinical collected signals.Based on the analysis of the amplitude-frequency and time-frequency characteristics of fetal ECG,maternal ECG and noise interference,the simulated ECG signals are generated by using MATLAB software,which are closer to the original ECG to verify the extraction effect of this algorithm.Firstly,the original signal is pretreated,comb filter is used to suppress power frequency noise,low-pass filter is used to eliminate EMG high frequency interference,and multi-layer wavelet decomposition is used to remove baseline drift in low frequency band.The pretreatment effect is optimized by experiments.Then,the mother’s ECG and noisy fetal ECG signals are separated by using the constructed FastICA algorithm model based on negative entropy.Then the noisy fetal ECG data are processed by time-frequency transform to obtain the non-negative spectrum,and the characteristic signals of fetal ECG are obtained by NMF decomposition.Finally,the position of R wave is located by peak detection method,and the fetal heart rate is calculated.The experimental results are compared and analyzed according to the evaluation index of the algorithm.Experiments show that the proposed method based on independent component analysis and non-negative matrix decomposition can extract clear mother ECG and accurate fetal ECG characteristic signals,and verify the feasibility and effectiveness of the algorithm.The test results have great value of clinical medical information,which can be used as a reference for medical staff to judge whether the fetus is distressed or not.