Selection of Methods Predicting Fetal Weight Using Ultrasonography

Jun Zhu

1999-01-01

Abstract:Objective The neural network and regression methods were used to compare the accuracy of predicting fetal weight. Methods Of a total of 390 cases, ultrasonographic parameters as BPD, HD, ThD, AD, FL, LL, STT and SCT were measured at 72 hour before delivery. Among them, 250 cases utilized training neural network systems and establish a series of regression formulae for fetal weight prediction. In the other 140 cases, fetal weights were by neural network, regression equation and the traditional method for comparison of accuracy. Results The 6 variable neural network has the highest coincide rate, 88.6%, with a relative error of 0.8±5.2%. The regression methods have a little lower coincident rate. Among them, the 6 variable regression equation has a coincident rate of 85.7%, and the 2 variable has a rate of 83.6%. AD, FL and BPD are useful variables which affect the accuracy in the fetal weight prediction. Conclusions 1. The neural network is a new method to predict fetal body weight with high accuracy. The regression equations, with their ease of use and preciseness, are still the mainstay of predicting fetal body weight as well. 2. AD, FL and BPD are important parameters in predicting fetal weight. 3. It is necessary to use different method in predicting different fetal weight, such as over weight and under weight.