Research On The Method Of Fetal Heart Rate Passive Detection Based On Convolutional Neural Network

The abdominal fetal electrocardiogram(ECG)signal plays an important role in the detection of fetal heart rate during pregnancy,which is beneficial to the early diagnosis of fetal health.The electrocardiogram monitoring of abdominal fetus is a passive non-invasive method,which is safe and non-invasive.However,it is very difficult to detect fetal heart rate from the abdominal ECG signals,mainly because the ECG signals collected from the maternal abdomen are inevitably interfered by various noises,among which the maternal ECG signal is the main interference,and their amplitude is far greater than that of the fetal ECG signals.In order to obtain a reliable fetal heart rate,the accurate location of fetal QRS complex is mainly obtained from the maternal abdominal ECG signals.Although domestic and foreign scholars have proposed a large number of methods to detect QRS complex in adult ECG signals,the analysis of non-invasive fetal ECG signals is still challenging,the main reason is that the current extraction accuracy of fetal ECG detection technology is still not high and the reliability is difficult to ensure.As a new and rapidly developing technology in recent years,deep learning can effectively help research on target detection and pattern recognition,and has become a hot topic in the field of signal and information processing.Among them,the convolutional neural network has been developed rapidly in the past decade and attracted more and more researchers’ attention.The biggest characteristic of convolutional neural network is that it is good at extracting high-level abstract features from raw data and has good generalization ability,so it is widely used in the scientific research fields related to feature extraction.In this paper,on the basis of summarizing the existing methods at home and abroad,the convolutional neural network was used to detect the fetal heart rate.The main research contents are as follows:(1)One dimensional convolutional neural network was used to detect the fetal QRS complex.This method firstly selects four channels of fetal ECG signals from the abdomen on Physio Net as the experimental data set,then evaluates the signal quality through the sample entropy,and excludes the channels with poor signal quality.A 50 Hz notch filter and a 5～100Hz band-pass butterworth filter were used to remove power line interference,baseline drift and high-frequency muscle myoelectricity interference.Finally,a 9-layer one-dimensional convolutional neural network was used to conduct training,verification and testing without cross by 100 ms segments.The results showed that the overall sensitivity of fetal QRS complex detection was 81.97%,the positive predictive value was 77.90% and the F1 value was 79.84%,and the overall accuracy of convolutional neural network detection was 79.74%.(2)Two-dimensional convolutional neural network was used to detect fetal heart rate.In this method,the amplitude of four-channel abdominal fetal ECG was firstly normalized to the interval of [-1,1],and then the one-dimension ECG sequence was converted into a two-dimension time-frequency map by short time Fourier transform in 250 ms continuous segments.Finally,the 11-layer two-dimensional convolutional neural network was used for cross validation and detection.The experimental results showed that the overall sensitivity of fetal heart rate detection was 87.56%,the positive predictive value was 86.94% and the F1 value was 87.25%,and the overall accuracy of convolutional neural network detection was 88.06%.