Study On The Cardiotocography Data Extraction And Feature Analysis And Fetal Status Assessment Methods

Cardiotocography(CTG)can be used to monitor fetal status continuously.Dynamic changes in two physiological signals,say the fetal heart rate(FHR)and the uterine contraction(UC),and CTG has become an indispensable auxiliary examination in obstetrics.Currently,the correct identification of category II electronic fetal monitoring images is the main difficulty physicians face.These images frequently appear in clinical practice and do not have typical features.In order to reduce the impact of subjective physician discrimination and the lack of uniform criteria for clinical CTG parameter interpretation,research was opened to automate the classification of fetal heart monitoring data.Since electronic fetal heart monitoring devices do not provide offline data channels and there is a lack of available public databases,researchers need to extract FHR and UC signals from CTG images.In response to the above research background,the research in this paper is divided into four main areas as follows.(1)An effective CTG signal data extraction method was proposed for the binary CTG image with complex backgrounds.Firstly,the background grid and special marker noise of the image are removed using the templates;then,the region where the signal curves are located and the starting and ending positions are localized using a projection map;subsequently,the data of each curve are scanned and extracted column by column,and the missing values are filled in with the full-variance filtering algorithm;finally,the extracted signals are amplitude corrected.Firstly,the experiments used the data in the open CTU-UHB database to simulate the generation of CTG images.Tests were conducted on these 552 simulated CTG images.The correlation coefficient,mean absolute error,root mean square error,and different domain features of the extracted FHR and UC signals with the original signals were compared and analyzed.Meanwhile,the quality of data extraction was verified on 293 clinical authentic CTG images.The extracted signals were compared with the original signals regarding curve change trends,and clinical features were also extracted for comparison,which verified the method’s effectiveness in this paper.(2)New feature parameters were extracted for the FHR signals extracted from the category II electronic fetal monitoring image.Firstly,the extracted FHR signals are pre-processed to remove the effects of long-missing values and unstable segments,and interpolation and length unification operations are performed;subsequently,the signals are decomposed and clustered using discrete convolutional wavelet transform and K-Means clustering algorithm;finally,three new feature parameters are extracted and normalized.(3)Automated classification using the extracted feature parameters of the FHR signal was performed to assess the fetal status.Three classifiers,Support Vector Machine,K-Nearest Neighbor,and Random Forest,were used to perform parameter search and build classification models for classifying fetuses into normal or abnormal states,respectively.The classification model established by Random Forest with optimal parameters had a recall of 82.46%,accuracy,precision,F1 values over 70%,and specificity of 60.17%.(4)All the research was integrated into the MATLAB GUI system,and the fetal status evaluation system with a simple interface and easy operation was designed.This study realized data extraction,FHR signal feature extraction,and automated classification of fetal heart monitoring atlas for fetal status evaluation,verified the feasibility of fetal status evaluation based on category II electronic fetal monitoring images,and provided a basis for improving the accuracy and clinical adaptability of CTG computerized automated analysis.