Estimation Of Fetal Weight Based On Artificial Neural Network

The ultrasonic estimation of fetal weight before delivery is very important on obstetrical clinic, especially in making sure whether the fetal belongs to IUGR, LGA or normal fetal, choosing delivery scheme and reducing complications. Since the end of 1970's, the measurements by ultrasound had been applied to obstetrical department. Measuring fetal biological parameters by ultrasound is a very important method to estimate fetal weight. Some researchers have done large of investigation on estimation of fetal weight, most part of them making use of statistic means, namely regression analysis. But veracity of regression equation is low and having some disadvantage of unreliability. In recent years, artificial neural network has been widely applied to resolve complicated medical information, accordingly processing analysis,consequence and forecast. In this study, the back propagation(BP) neural network was used to estimate fetal weight instead of regression equation. At first, 109 case data of stylebook acquired from functional department of the Second Huaxi Hospital were variously analyzed by using regression. It has been found that four regression equations have better accord rate, even achieving 75%. The parameters of the equations describe the course of fetal upgrowth: including fetal head,limb,abdomen,fattiness and glycogen. In experiment of applying the back propagation neural network to research those stylebooks, rather ideal results of estimation have been achieved by three testing methods. All accord rates can reach 80%, the top being 95%. In grouping of training and validating, the accord rate is 89.77% and 76.19%; averagely absolute error is 104.22g and 190.84g; averagely relative error is 3.24% and 5.6%. It shows that neural network is more validate than regression. Moreover, the ultrasound parameters have been compared to find out the best parameter combination. GA is a valuable parameter when GA is measured in day rather than week in company with other ultrasound parameters. At last, the software based on Visual C++ is developed for clinic fetal prediction using back propagation neural network. The study is the base of making prediction equations more accurate and reform the criterion of fetal growing.