Research On CTG Signal Digitization And Intelligent Analysis Algorithms For Fetal Monitoring

During the perinatal period,Cardiotocography(CTG)is the most popular prenatal diagnosis.In clinic,the output form of fetal heart contraction monitoring is mainly paper-based CTG report,but its visual analysis and interpretation are lack of objectivity and repeatability,and the sharing database is also extremely lack,so a set of widely applicable digital tools is needed.In order to help doctors diagnose accurately,it is very important to use automatic technology to monitor pregnant women and fetus.As a main parameter in clinical detection,uterine contraction has a high value in the research of intelligent system.In this paper,CTG signal digitization and intelligent classification algorithm of different intensity uterine contraction signals for fetal monitoring are mainly studied as follows:(1)This paper proposes a digital algorithm for binary CTG paper reports,which breaks through the limitations and discomfort of existing algorithms for binary CTG paper reports.Firstly,the CTG image is acquired by smart phone.In the stage of gridline removal,an area fusion method based on super pixel is proposed,and an improved binary line mask method is designed.The connected area is used to remove the gridlines and dashes which are independent of the signal lines perfectly.Secondly,in the signal extraction stage,according to the different states of each signal trace,different methods are used to extract the representative pixels.Finally,B-spline is used to fit the extracted signal and interpolate to fit the actual signal.The amplitude and time of the signal are corrected by horizontal and vertical histogram projection respectively to realize the time synchronization between the interpolation signal and the actual signal.At the same time,a known database was used to evaluate the digitization effect from both quantitative and qualitative aspects.(2)For the acquired digital uterine contraction signal,an intelligent uterine contraction classification algorithm based on multi-dimensional feature extraction is constructed,which realizes the idea of computer-aided method to help doctors accurately diagnose.In signal preprocessing,empirical mode decomposition combined with morphological filtering is proposed to remove high frequency noise,and smooth prior method is introduced to remove baseline drift noise.In the construction of eigenvectors,using the recursive analysis strategy,a multi-modal eigenvector is designed,which combines one-dimensional time feature and two-dimensional recursive feature.Finally,the optimized SMOTE-PCA-SVM classifier is used to classify the intensity of contractions intelligently.In this paper,the signal-to-noise ratio and other indicatorsare introduced to analyze and compare the performance of the designed filtering algorithm and different machine learning classifier algorithms.The research in this paper realizes the digitalization of CTG signal and the automation of uterine contraction signal,which lays a theoretical foundation and technical support for the application of CTG fetal monitoring.