Research On Fetal ECG Extraction Based On Improved ICA And Wavelet Transform

Fetal ECG is the source signal of heart activities of fetus, and directly reflect healthy condition of fetus during pregnancy. The abnormal phenomenon during fetal growth process could be found by the inspection of fetal ECG and FECG inspection is also a powerful method to monitor fetal safety condition in uterus. Focused on the problem of extraction of FECG, an improved FastICA algorithm based on damped Newton iteration is proposed.Based on the research, analysis and comparison on the current methods for extraction of FECG, it is discovered that the numbers and precision of samples have important influence on the results of the extraction of FECG. Good performance of extraction of FECG usually requires high numbers and precision of samples, which will bring long length of data and increase the complexity of calculation. So an algorithm with good performance and efficiency should be found to extract FECG. The research shows that the method based on ICA has advantages of good performance and the capability of independent non-Gaussian among components is satisfielded. This kind of method develops rapidly and has a wide range of applications. So the research of this thesis focuses on FastICA algorithm, and proposes a new method for extraction of FECG based on improved FastICA algorithm. In this thesis, damped Newton iteration is used instead of the Newton iteration in original FastICA algorithm, and one dimension search is imposed on the direction of Newton iterative in order to overcome the drawbacks, which the choice of initial value is sensitive. It increases the convergence speed and improves the quality of the extracted signals. The experiments based on synthesized and real signals demonstrate that this method has more robustness and adaptability for extraction of FECG. The extracted signals are more clear and the convergence speed of algorithm is faster.Since the signals extracted from actual ECG by improved FastICA still contain some noise, the wavelet threshold denoising is utilized to denoise the extracted signals for further processing. An improved method for threshold denoising is presented. A new threshold function is constructed. It syncretizes traits of soft-threshold and hard-threshold denoising methods by weighted average and the parameters can be adjusted properly to produce the best estimations of the wavelet coefficients. The accuracy of reconstruction is improved. The experimental results show that the quality of the denoised ECG is significantly improved. The detail of ECG is more clear, and bring convenience for the clinical analysis. At last, the summary on this research is described and the prospect of FECG extraction is also proposed.