Recognition Of Uterine Contractions With Electrohysterogram Signals

Uterine contraction can prompt the fetus to be excluded from the mother.Labor is defined loosely as regular uterine contractions accompanied by cervical effacement and dilation.With the approaching of delivery,uterine contractions will become more frequent and synchronized.Preterm labor is defined as the labor before 37 weeks.And uterine contraction and cervical length are important basis for clinical diagnosis of preterm delivery.In 2010,the global preterm birth rate was about 10%,and preterm birth has become the second leading cause of death for children under five years of age,only after pneumonia.Therefore,the effective monitoring of uterine contraction is of great significance to the health of pregnant women and fetuses.The study mainly included as follows:First of all,the study of characteristics extraction method of Electrohysterogram(EHG)between uterine contraction and non-contraction.The characteristics mainly include time-domain,frequency-domain and non-linear characteristics.And Mann-Whitney U test was employed to analyze the significance of the characteristics between uterine contraction and non-contraction.Secondly,traditional machine learning algorithms:Support Vector Machine(SVM),Decision Tree(DT),Artificial Neural Network(ANN)and deep learning algorithm:one-dimensional Convolutional Neural Networks(1-D CNN)are used to recognize EHG segments of uterine contraction and non-contraction.The performance of the classifiers which were trained by different kinds of characteristics were compared.The ability of characteristics to recognize contraction and non-contraction EHG segments was also ranked.The performance of the classifiers were evaluated by sensitivity,specificity,PPV,NPV and Accuracy.The study showed that the mean square root(RMS),standard deviation(STD),mean absolute value(MAV),simple square integral(SI),difference absolute standard deviation value(DAS),average amplitude change(AAC),peak frequency(PF),energy and time reversibility(TR)extracted from contraction EHG segments were significantly higher than those extracted from non-contraction EHG segments(p<0.05);And the median frequency(MF),sample entropy(SamEn)extracted from contraction segments were significantly lower than those extracted from non-contraction segments(p<0.05).There are no significant differences in Lyapunov Exponent(LE)and LOG detector(LOG)between contraction and non-contraction(p>0.05).The classification results of various classifiers used in this study are as follows:For SVM,the SVM trained by all characteristics of channel 1 performed best AUC=0.79,sensitivity=0.88,PPV=0.77,NPV=0.86,accuracy=0.81;For DT,the DT trained by all characteristics extracted from all channels performed best AUC=0.81,sensitivity=0.80,specificity=0.80,PPV=0.80,NPV=0.80,accuracy=0.80;For ANN,the classifier trained by the non-linear characteristics of channel 1 performed best with AUC=0.79,sensitivity=0.91,specificity=0.69,PPV=0.75,NPV=0.89,Accuracy=0.80;CNN showed AUC=0.73,sensitivity=0.68,specificity=0.83,PPV=0.88,NPV=0.58.So in this study,DT is the best method to identify contraction and non-contraction EHG segments.This study researched the changes of characteristics between contraction and non-contraction status.The classifier studied in this paper can effectively identify contraction and non-contraction EHG segments.This study proposed an accurate method of recognizing uterine contraction,which can provide an alternative method for doctors to diagnose preterm birth and other abnormal conditions.