Two-stage Extraction Of Fetal ECG Signals Using Single-channel Maternal Abdominal Signals

Fetal ECG signal is an important means to monitor the fetal health status during the perinatal or pregnancy period.It can effectively find the unhealthy development and growth of the fetus in the mother and reduce the morbidity and mortality of the fetus,Therefore,extracting a clear fetal ECG signal has important theoretical significance and practical value.Due to the complexity of the fetal growth environment in the mother,the fetal ECG signal collected is very weak,and will be interfered by factors such as the mother’s ECG signal,power frequency interference,baseline drift,etc.,making it difficult to effectively extract the fetal ECG signal.Non-linear estimation models have been widely used in the extraction of fetal ECG signals,but the existing methods usually require more channels of ECG signal data,which increases the complexity and difficulty of the model.Moreover,the hyperparameters in the model rely on experience to obtain values,which directly affects the predictive performance of the model,making it difficult for the model to extract a clear and complete fetal ECG signal.In view of the above situation,this paper combines the singular value decomposition technology(SVD)and the least square support vector machine(LSSVM)optimized by the cuckoo search algorithm(CS)to propose a new two-stage fetal ECG extraction method based on single-channel maternal abdominal wall mixed signals.This method only needs to record the pregnant abdominal wall mixed ECG signals once,and can automatically search for the optimal value of hyperparameters in the model.In the first stage: firstly,the smooth window technology(SW)is combined with SVD(SW-SVD)to estimate the maternal ECG signal components in the mixed ECG signal of the pregnant woman’s abdominal wall;then,take the maternal ECG signal estimated by SW-SVD as the input,and take the single-channel mixed signal of pregnant woman’s abdominal wall as the target output,use LSSVM to construct a nonlinear mapping model to obtain the best estimate of the maternal ECG signal in the abdominal wall mixed signal,and search for the optimal hyperparameters in the model through the CS algorithm;finally,the mixed signal of the abdominal wall is subtracted from the best maternal ECG signal estimated by the model,and the preliminary fetal ECG signal can be separated.In the second stage: the SW-SVD technology is used again on the fetal ECG signal initially obtained to further eliminate the interference of the maternal ECG signal and other noises,and finally obtain a pure fetal ECG signal.The proposed method is analyzed experimentally using two measured ECG signal databases,and the results are evaluated in terms of visualization results and error performance indicators.Experimental results show that the method proposed in this paper can extract a clear and complete fetal ECG signal more accurately,and has a better extraction effect than the classic fetal ECG extraction model.