Cyclostationarity-based Semi-blind Extraction Of Fetal Electrocardiogram

Fetal Electrocardiogram (FECG) is a physiological signal that originates from fetalcardiac beat, transmits via the uterus to the outside and finally can be detected in thematernal abdomen wall. Complete FECG signal can distinguish the slight variation incardiac potential accurately and may be an important tool for diagnosing the fetalcardiac defects. However, due to the overlap in time domain and frequency domain ofthe maternal electrocardiogram (MECG) and FECG, relatively low signal-to-noise ratio(SNR) and non-stationary of the FECG, shortage of a priori in early studies,non-invasive FECG extraction and clinical FECG application is very difficult.In this study, a cyclostationarity-based semi-blind method is proposed for FECGextraction. Theoretical analysis shows that this method can remove the remained MECGinterference and muscle artifact. The main work and innovation points are as follows:①In composite abdominal signal，MECG has a quasi-periodic structure and Rwave can be easily detected. To remove the MECG interference channel by channel, asingle channel SVD method based on the principle component analysis (PCA) isproposed. Firstly, constructing a periodic matrix with each row contains a MECGheartbeat; secondly, using SVD to extract the principle component from the periodicmatrix; lastly, reconstructing the single channel signal, eliminating the principlecomponent from the original abdominal signal and most MECG interference isremoved.②FECG is a cyclostationary signal. To extract the FECG from remainedcomposite signal, a cyclostationarity-based semi-blind extraction (CBSBE) methodbased on the generalized eigenvalue ICA is proposed. Firstly, estimating the cyclicfrequency of FECG and constructing the separation rule based on the degree ofcyclostationarity; secondly, transforming the separation rule to the form of generalizedRayleigh quotient; lastly, maximizing the separation rule to extract the FECG andremoving the remained MECG interference and muscle artifact.③Combining the proposed single channel method and multichannel method, anovel semi-BSS system for FECG extraction is proposed. The method consists of fourparts: preprocessing, pre-whitening, semi-BSS and post-processing. The idea of theproposed is removing the MECG and noise sequentially and extracting the clear FECGlastly. Results show that the proposed semi-BSS system can extract clear FECG andestimate fetal heartbeat and FECG wave more accurately as compared with theconventional source separation methods. Moreover, results also show that the proposedmethod can extract clear FECG signal while the lead system is in the sameconfiguration.