Study On Quantitative Analysis Of Cardiotocograph And Fetal Condition Assessment Model

Cardiotocography(CTG)is an important output form of electronic fetal monitoring records.Through CTG can analyze parameters including fetal heart rate baseline,acceleration,deceleration,and the relationship between uterine contraction and fetal heart rate deceleration,reflecting the fetal health status.Relying on doctors for manual interpretation is time-consuming and subjective.Using computer automatic analysis of CTG can assist obstetricians to make clinical decisions to a certain extent.However,the lack of fetal heart contraction curve data and noise,as well as the complexity of fetal heart rate patterns make it difficult to analyze and assess fetal status.Therefore,this paper studies to improve the accuracy of quantitative analysis of the characteristics of fetal heart contractions,and attempts to solve the problem of fetal state assessment.The research content of this article is as follows:(1)Research on Extraction Algorithm of Contraction Curve.This paper designs the method from two angles of contraction baseline extraction and contraction wave identification.Improve the contraction baseline algorithm and use the method based on the minimum mean square error of the sliding window distribution interval.This method can effectively follow the contraction baseline change and other conditions.Design a contraction recognition method,set criteria in terms of peak value,duration,baseline change,and morphological characteristics of uterine contractions to distinguish between contractions and pseudo contractions and other interferences Perform algorithm verification experiments on clinical data to obtain 91.8% sensitivity and 90.8%positive predictive value.(2)Investigation and research on the algorithm of fetal heart rate signal baseline and deceleration feature extraction.Aiming at the previous problems in the extraction effect and accuracy of the baseline extraction and deceleration recognition algorithm,the baseline algorithm for fetal heart rate based on the moving window rms reference value is improved.Compared with the other seven different baseline extraction algorithms,the average absolute error of the baseline algorithm is 7.5%.Two algorithms are proposed to identify and locate the fetal heart rate deceleration zone(The following article will be referred to as the deceleration zone): one is the deceleration zone recognition algorithm based on the iteration threshold.The second one is to classify the original signal point by point based on one-dimensional shallow U-net to get the result partition.Experiments were conducted on clinical data with expert logos and fetal heart rate data from the Catholic University of Lille,France,and the results marked by the experts were used as standard references and compared.The experimental results show that the average relative error of the iterative threshold deceleration zone identification method in this paper is reduced by 6.1% compared with the traditional method;the U-net network positioning and recognition results for the deceleration zone are also ideal.The sensitivity and specificity were 88.84% and 89.34% respectively.The above two deceleration zone identification algorithms are equipped with computer hardware environments Different requirements.Next,the integral method is used to calculate the area of the deceleration area(FHRDA).Compared with the calculation result of the traditional approximate triangle method,the calculation error of the algorithm / approximate triangle method for the V / U / V + U type deceleration area is 2.2% / 6.2%,4.4% / 10.3%,29.7% / 76.8%,compared with the Vshaped deceleration zone,the calculation accuracy of the U-shaped or U / V mixed deceleration zone is higher.The results were evaluated with p H as the gold standard,After non-parametric single-factor analysis of variance,the area obtained by the effective integration algorithm is significantly different from that of fetal acidemia;then iterative threshold and traditional one are used to identify the deceleration zone and the integration method is used to perform cross-comparison experiments,respectively,highlighting the method Effectiveness,the experimental results show that the sensitivity(62.1%)and specificity(80.0%)of this method are higher than the sensitivity(52.3%)and specificity(74.8)of the traditional method when the deceleration area of the fetal heart rate curve reaches 8.5 cm2 %)has seen an increase.(3)Study on the non-linear characteristic parameters of fetal heart rate signal to characterize acidemia.Compare the differences between the parameters and the effects of different machine learning methods on the evaluation results of fetal acidemia.Try to convert the fetal heart rate signal into a two-dimensional image for classification and analysis of acidemia.Using the convolutional neural network as the method of evaluating the model,the one-dimensional signal is converted into a two-dimensional recursive graph as the input of the training network,p H = 7.2 is used as the classification reference limit standard to evaluate the experiment,and the model is evaluated using 10-fold cross-validation and after adjusting the parameters,the classification accuracy is 98.38%.