Research Of Fetal ECG Extraction Method From Abdominal Maternal ECG

Heart defect is the most significant factor in the normal birth and the fetal death.According to statistics, about one out of125babies is born with some form of congenitalheart defects every year. Fetal health monitoring is very important during the pregnancy. Atpresent, the health of the fetus monitoring is mainly based on heart sound and heart rate,while the Electrocardiogram (ECG) can response to the subtle cardiac electrical activitywhich has extremely important meaning and function for the diagnosis of heart disease. Webelieve that fetal electrocardiogram (FECG) provides more valuable clinical information ofits physiological state than heart sound and heart rate. So we hope to get a clear fetalelectrocardiogram.Noninvasive measure methods use the signals recorded from the maternal abdominalwall, they can be done in any stage of pregnancy using dozens of electrodes. Noninvasivemeasure method is to put the electrode on the maternal abdominal wall which can measurethe fetal electrocardiogram in any stage of pregnancy. However, the abdominalelectrocardiogram (AECG) have low fetal ECG SNR and it contains many noise, such asbaseline wandering, power line interference, electromyogram (EMG) and maternalelectrocardiogram (MECG). The most interference factor derives from the MECG, hence,our work is MECG cancelling.The algorithm of extracting the FECG is divided in many kinds of algorithms: Filter,Machine Learning, Linear Decomposition and other algorithms. This dissertation choosethree methods with the characteristic of real-time, efficient and stability from these threealgorithms respectively: LMS Adaptive Noise Cancellation, RBF networks, In IndependentComponent Analysis.First, using these three algorithms to extract the fetal electrocardiogram based on theDaISy database.And compared the extraction effect of each algorithm from two aspectswhich were the visual waveform observation and SNR (signal-to-noise ratio). Thisdissertation successfully extracted fetal ECG signals by using these three algorithms fromabdominal wall mixed signal and chest mother ECG.But the extracted fetal electrocardiogram contains a small amount of the mother ECGinterference and the residual noise when using the first two algorithms compared by thewaveform observation. The independent component analysis algorithm is superior to thefirst two algorithms obviously and the calculation of noise ratio also proves this result. Then, to verification on the FastICA algorithm using the MIT database, this dissertationemploys a low pass first order Butterworth filter and IIR notch filter to remove power lineinterference and gets the four channel abdominal mixed signal with relative small noise.And fast independent component analysis is applied to the four channel signal, ultimatelythe fetal ECG signal with high SNR is successfully extracted.