Research Of Fetal State Intelligent Assessment Algorithm Based On CTG Parameters

In recent years,with the comprehensive implementation of the "two-child" policy and the widespread popularization of the concept of the "bear and rear better children",people’s awareness of safeguarding maternal and child health has gradually increased.Real-time monitoring of the fetal state helps clinicians detect and treat fetal abnormalities in a timely manner,thus reducing the rate of birth mortality and infant defect.Cardiotocography（CTG）is the most common and most effective method of fetal monitoring in the diagnosis of intrauterine health,among which the fetal heart rate（FHR）signal is a kernel parameter of perinatal electronic fetal monitoring（EFM）.The interpretation of CTG signal is essential to timely,effective,accurate,and convenient access to the health status information of pregnant women and fetuses contained in CTG,which has important theoretical value and practical significance for improving the birth quality of the population.The experience of medical staff in interpreting CTG signal is uneven,and the inconsistent clinical diagnosis results often happen due to subjective differences,thus promoting the rapid development of computer-aided analysis system.However,the current computer-aided systems still have many shortcomings,such as insufficient parameters and limitations of algorithms,which make the accuracy of diagnosing fetal status is not high.The thesis mainly studied the fetal state intelligent assessment using the artificial intelligence（AI）learning algorithm.First,the multi-modal parameters based on CTG signal were extracted.On this basis,the feature optimization method was used to select the optimal feature subset,and then the machine learning（ML）was constructed to perform fetal classification.Finally,without the steps of feature extraction,the convolutional neural network（CNN）structure was designed and optimized to make a diagnosis of the fetal state.The main works of this thesis are organized as follows.（1）The research background and significance of this thesis were explained.The research status of fetal state assessment algorithms at home and abroad were reviewed,which provided the theoretical basis for the research scheme proposed in this thesis.（2）The multi-modal parameter extraction algorithm based on CTG signal was proposed,and the visualization software was developed.The comprehensive parameters of different feature domains from preprocessed CTG signal were extracted,including:the morphological features which describe the morphology of the curve,the time domain features which reflect the minor changes of the FHR signal,the frequency domain features which calculate the power spectral density of the signal,and the nonlinear features which represent the nonlinear characteristics of the signal.And the computer software for visualizing the analyzed results was designed.（3）The fetal state evaluation algorithm based on ML was proposed.Based on the extraction of CTG parameters,the synthetic minority oversampling technique（SMOTE）was used to alleviate class imbalance problem.The method applied to select the optimal feature subset from the original dataset was studied,and several machine learning algorithms were designed to execute the fetal classification task,and the performance of the algorithm was tested using the open-access database.Finally,the experimental results showed that the performance of classification based on genetic algorithm（GA）and adaptive boosting（AdaBoost）was the best,which achieved the accuracy（ACC）,sensitivity（SE）,specificity（SP）and area under the curve（AUC）of 95.15%,96.51%,93.44%and 94₆7%,respectively.（4）The fetal state evaluation algorithm based on CNN was proposed.The continuous wavelet transform（CWT）and recurrence plot（RP）were applied to convert the one-dimensional FHR signal into two-dimensional image.The structure of the two-dimensional CNN was designed and optimized.The training strategy method was cross-validation and the performance of the algorithm was tested using the open-access database.Finally,the experimental results showed that compared with the traditional ML method,the CNN can self-learn relevant information from the input without the process of feature extraction and optimization,and the classification performance was better,which achieved the ACC,SE,SP and AUC of 98.36%,99.05%,98.55%and 98.36%,respectively.The research of this thesis realizes the automatic monitoring and analysis of FHR parameters and the intelligent assessment of fetal health status,which provides the theoretical foundation and technical support for the engineering application of the clinical perinatal fetal intelligent monitoring instruments.