Research On Fetal ECG Signal Extraction Method Based On Temporal Convolution Encoding And Decoding Network

Fetal electrocardiogram(FECG)monitoring is currently the most promising fetal health monitoring method.Its acquisition method is mainly to record the electrophysiological signals of fetal heart activity by placing electrodes on the abdominal wall of pregnant women.Similar to the application value of adult ECG signals,the waveform characteristics of fetal ECG signals can be used to obtain important diagnostic information such as fetal heart rate and physiological metabolism,which can assist clinics to more accurately judge whether the fetus is healthy or not through heart rate and ECG waveforms.Therefore,it is of great significance to extract fetal ECG signals with clear waveform features.However,the abdominal electrocardiogram(AECG)signal collected from the abdominal wall of pregnant women is a mixed signal,and it is a difficult to obtain the weak fetal ECG signal from it.There are two main reasons:(1)The energy of the maternal electrocardiogram(MECG)component in the AECG signal is much greater than that of the FECG;There is overlap in the above,which makes the extraction of FECG signal extremely vulnerable to the interference of MECG.(2)AECG signals are often mixed with various noises,such as maternal electromyogram(EMG),baseline wander(BW)and random noise.Existing related research is still limited to extracting fetal ECG and detecting the position of R-peak to estimate fetal heart rate,and the complete extraction of fetal ECG waveform has not yet been realized,which greatly limits the clinical application value of fetal ECG signal.In this thesis,combined with adaptive noise cancelling(ANC)framework,a fetal ECG signal extraction method based on temporal convolution encoding and decoding network(TCED)is proposed,which realizes the extraction of fetal ECG signals from pregnant women.The method is mainly divided into two stages:(1)The elimination of the abdominal maternal ECG component.After completing preprocessing such as baseline drift correction and high-frequency random noise filtering by using the complete ensemble empirical mode decomposition with adaptive noise,the pregnant woman’s thoracic ECG signal was taken as the input,and the synchronously collected abdominal ECG signal was used as the mapping target.The TCED estimated the best abdominal wall maternal ECG component and subtracted it from the abdominal ECG signal to obtain the residual signal including fetal ECG.(2)The denoising of the fetal ECG signal.After eliminating most of the maternal ECG component in the abdominal ECG,the TCED was used to denoise the extracted multi-channel fetal ECG signals to further eliminate maternal ECG and random noise residues and improve the signal-to-noise ratio of fetal ECG signals.Finally,in the data post-processing link,based on the extracted multi-channel fetal ECG signals,we use the independent component analysis(ICA)separation algorithm to obtain singlechannel fetal ECG signals with complete waveforms and clear features.We use simulated ECG data(FECGSYN)and real ECG data(NIFECGDB,PCDB)to test and compare the performance of the proposed method.The results show that the average fetal ECG R-peak detection sensitivity(SE),positive predictive value(PPV),and F1 score of this method in FECGSYN reached 98.47%,98.97%,and 98.72%;mean square error(MSE),signal-to-noise The ratio(SNR)and correlation coefficient(r)reached 0.16,7.94,0.95;compared with the average performance index of EKF(SE=92.36%,PPV=94.60%,F1 score=93.45%,MSE=0.38,SNR=4.26,r=0.86)were higher by 6.11%,4.37%,5.27%,0.22,3.68 and 0.09,respectively.The average fetal ECG R peak detection sensitivity(SE)and F1 score in NIFECGDB reached 99.02%and 98.94%,which were 0.99%and 0.99%higher than the average performance index of GFLANN(SE=98.03%,F1 score=98.47%)respectively.0.47%;in terms of positive predictive value(PPV),the two were basically the same(98.86%vs 98.92%).In PCDB,the average fetal ECG R-peak detection sensitivity(SE),positive predictive value(PPV),and F1 score reached 98.61%,98.63%,and 98.62%,which were higher than the average performance indicators of AECG-DecompNet(SE=93.52%,PPV=97.41%,F1 score=95.43%)were 5.09%,1.22%,and 3.19%higher than that,respectively.The results show that the method in this thesis can extract a clearer fetal ECG signal,which is helpful for the analysis and diagnosis of the fetal ECG waveform,and has certain application value for effective fetal health monitoring during pregnancy.