Robust Suppression Of Maternal Electrocardiogram Interference In Noninvasive Fetal Electrocardiogram

Fetal electrocardiogram (FECG) could reflect very slight variations of originalelectrical activity of heart, so has broader medical prospects than fetal heart sounds andmagnetic monitoring. At present, FECG is obtained mainly through gravida abdominalwall detection, which faces more complex noise environment in addition to the maternalECG (MECG) interference because of maternal breath, contraction and quickening andso on. All of these bring much trouble to the FECG’s extraction.In this paper, the MECG suppression method was adopted to suppressMECG andreserve the FECG as much as possible, so as to get stable and reliable fetal heart rateand complete FECG. By combining the stable and quasi-periodic properties of MECG,here the robust model-based M-estimation and particle filter were introduced tosuppressMECG.Firstly, abdominal MECG(AMECG) was preprocessed and model’s initialparameters were obtained. Preprocessing included removing baseline drift and powerline interference with fast and stable methods, and building the statistic model of musclenoise and other impulse noises. Model’s initial parameters of MECG were obtained byAMECG’s feature point detection and average heart beat fitting with M-estimation inthis paper.Then, static model of MECG was established and M-estimation was presented forMECG extraction. Pulse noise and FECG in preprocessed signals had obviously pulsecharacter, which lead traditional estimation theories based on Gaussian statisticaldistribution were not suitable, so robust M-estimation was presented for MECGextraction in this paper. And in order to improve the effect of estimation, classical3DGaussian kernel model was improved to double-hemi Gaussian kernel model. Throughthe simulation experiments among least square method and three M-estimate,M-estimation was verified more effectively than least square method for MECGsuppression, while under the condition of R-peaks overlap between FECG and MECG itwasn’t stable and the estimation error was large, so M-estimation was more suitable forestimating stable MECG.Finally, dynamic model of MECG was established and particle filter was presentedfor MECG extraction. Because M-estimation wasn’t stable when R-peaks overlapbetween FECG and MECG, and MECG was a typical discrete dynamic system, particles filter as a dynamic robust filtering method was presented in this paper to extract MECG,which local tracking ability was stronger than M-estimation. Combining with moreMECG’s prior, two state space model based on double-hemi Gaussian kernel modelwere deduced, namely2-state model and18-state model. Through the simulationexperiments between extended Kalman filter (EKF) and particles filter (PF), PF with2-state model was verified more effectively than other methods for MECG suppression,especially solved R-peaks overlap between FECG and MECG to some extent, and couldobtain more stable and accurate fetal heart rate and more complete FECG. But PF costmore times to estimation states, so suitable for the situation which real-timerequirements was not high.