Pattern Recognition Of Fetal Heart Function In Gestational Diabetes Mellitus

Introduction:compared with the normal fetal, gestational diabetes mellitus fetal ventricular wall and sepal thickening in pregnancy, cardiac systolic function increased, diastolic function decreased significantly, change about many indices of cardiac function, even the fetal cardiac morphology and function of satisfactory control of maternal glucose still have the change. In this study, changes in diabetic pregnancy fetal heart index model variables, respectively, using Logistic regression model, support vector machine (SVM) model, radial basis function network (RBF Network) model of machine learning method to filter out characteristic index of heart function in fetuses of diabetic pregnancy, provide a strong basis for the relevant decision-making. Methods: Based on a descriptive analysis, whether to have GDM as the dependent variable, with many cardiac structure and function index as dependent variable, through four kinds of feature selection method, the combined application of support vector machine (SVM), radial basis function network (RBF Network) and Logistic regression analysis of three kinds of classification methods, screening out the characteristic index cardiac function in fetuses of diabetic pregnancy. Results:the sensitivity of1Logistic regression model, SVM model, RBF Network model three kinds of methods of forecasting and specificity were75.9%and91.5%,78.8%and91.9%,79.6%,91.9%;2three methods of screening results varied (SVM model selected12indexes; RBF Network model selected9indexes; Logistic regression model selected13indexes), but ventricular septum thickness, three tricuspid annulus peak late diastolic velocity determined by these three methods were screened out;3of three classification methods of ROC (Receiver operating characteristic) area under the curve were AUC (SVM)=0.894, AUC (RBF Net)=0.940, AUC (Logistic regression)=0.927; P-R curve (Precision-Recall curve, precision and recall curves) area are AUC (SVM)=0.850,AUC (RBF Net)=0.903, AUC=0.880 (Logistic regression). Conclusion:in the classification of screening for gestational diabetes fetal cardiac structure and function in RBF Network, compared to SVM, Logistic regression showed better effect. Ventricular septum thickness, three tricuspid annulus peak late diastolic velocity is a characteristic index of heart function in fetuses of diabetic pregnancy.