Analysis Of Fetal Heart Rate Feature Extraction And Its Application In Electronic Fetal Monitoring

Currently,Electronic fetal monitoring(EFM)is most widely used in clinical obstetrics intrauterine monitoring and it is an important way to understand the intrauterine fetal situation and fetal reserve capacity.EFM has an important influence on predicting the occurrence of fetal distress or judging high-risk fetal’s situation.Nowadays,antenatal fetal heart rate(FHR)monitoring techniques mainly include ultrasonic Doppler method,magnetocardiogram and abdominal fetal ECG method.The advantages of ultrasonic doppler method are economy and simple operation which makes the method is widely used in clinic,while,the limitations of the doppler technique let the pregnant woman neither move,nor take long time monitoring(20 to 40 minutes),so it will ignore many important fetal potential information.With the aid of fECG fetal heart monitor,pregnant women can move freely and the monitoring time can prolong to 24 hours,which is the international forefront method of advanced fetal heart monitoring.Due to the low frequency prenatal contractions,analysis of FHR signal is the main means of understanding the state of the fetus and it can help reduce the maternal mortality rate,perinatal mortality and disabled children.Ultrasonic Doppler fetal heart monitor and the monitor based on the technology of fECG mainly get simple time domain parameters.This study extract FHR morphology,time domain and nonlinear parameter to analyze the differences between different types of fetus and FHR characteristic parameters’ trend with the change of monitoring time and gestational age,then select the optimal parameters.Finally,using machine learning methods to establish a classification model then classify the fetus.This study based on the technology of fECG maternal-fetal Holter monitor to gather mother heart rate,FHR and uterine contraction,then design algorithm to extract the baseline heart rate,acceleration,variation,sleep-wake cycle and nonlinear parameters.Using statistical methods to analyse continuous monitoring FHR’s various characteristic the mean and range of parameters.The results show that the baseline with a downward trend from10 p.m.to 4 a.m.and the lowest around 2 a.m.Normal fetal acceleration area and time significantly higher than the suspicious fetus.However,there was no significant difference of acceleration times between two groups.Normal group small variation ratio is lower than suspicious group while moderate variation ratio is higher than suspicious group.Besides,normal fetal quiet sleep time length is also less than suspicious group of fetus.This study extracts 36 FHR parameters,and there are 22 parameters have a significant difference between two groups.Regression analysis and machine learning are used to distinguish normal and suspicious fetuses.The regression equation’s predicted accuracy is 80.95%.Eighty percent of the 84 data was used to train a classification model and testing set of accuracy can reach 93.75%.