Research On The Methods Of Fetal Electrocardiogram Signal Extraction

The fetal electrocardiogram (FECG) could reflect the healthy condition of the fetus.The fetus healty condition could be monitored through the perinatal FECG monitoringand observation to reduce the morbidity and mortality rates. But the clear FECG wasdifficult to extract by the existing methods which limits the clinical application on alarge scale. In this paper, the research background and significance were presented andthe development of the FECG extraction methods was analyzed.Recently, blind source separation (BSS) method has been introduced to the FECGextraction field, which has been considered as a better and promising method. Due tothat the traditional BSS method was limited to the non-Gaussian or stable signal, andthe ability of noise inhibition was not strong. In this paper, FECG extraction based onTFBSS combined with EMD was presented. The proposed method was compared withRLS and DNN from visual results and quality assessments using real data. EMD wasused to denoise, and the denoising FECG was compared to the extracted FECG.Due to that the FECG extraction method based on BSS needs more leads and withhigh computation complexity. In this paper, the v-SVR was described in detail on thebasis of the statistiacal learning theory and traditional support vector regression (SVR).The steps of the fetal electrocardiogram (FECG) signal extraction based on v-SVR onlyusing two recordings was presented in detail. In this paper, the syntheticelectrocardiogram (ECG) signals generated by the dynamical model developed byMcSharry and the real ECG signals contributed by Lieven De Lathauwer were adoptedto illustrate the advantages of the performance based on v-SVR. The proposed methodwas compared with the other four FECG extraction methods: online least squaressupport vector machines (OLS-SVM), adaptive neuro-fuzzy inference systems (ANFIS),dynamic neural network (DNN) and recursive least square (RLS) from the visual resultsand quality assessments. The quality signal to noise ratio (qSNR) and the correlationcoefficient r were used for the synthetic ECG signals to analysis the performance ofextracted FECG, the eigenvalue analysis technique (SNRe ig) and the cross correlationtechnique (SNRR MS) and training time were used for the real ECG data.Finally, the extracted FECG based on v-SVR was denoised by EMD to obtain themore clear and high SNR signals. For real data, the extracted FECG based on v-SVRand the extracted FECG based on TFBSS were compared from the normalized visual results and quality assessments, the denoising FECG based on v-SVR and the denoisingFECG based on TFBSS were also compared from the normalized visual results andquality assessments.