Abstract: Objective：To compare the accuracy of traditional prediction method and back propagation artificial neural network (BP-ANN) for estimation of fetal weight (EFW), to explore the best model for EFW. Methods：A total of 224 term pregnant women with singleton between November 2015 and July 2016 in North China University of Science and Technology Affiliated Hospital were included in the present study. They were divided into non-macrosomia group (FW＜4 000 g, n=183) and macrosomia group (FW≥4 000 g, n=41). An ultrasound examination was performed 0~5 days before delivery. Clinical data and ultrasound data were collected. Results：Reduced errors were seen when measurements were obtained by ultrasonic parameter and combined parameter BP-ANN than traditional prediction method (P＜0.05). In macrosomia group, the error of combined parameter BP-ANN decreased with the increase of the number of training samples (P＜0.05). For BP-ANN, less error were seen when measurements were obtained by combined parameter than ultrasonic parameter and clinical parameters in non-macrosomia group and overall (P＜0.05). Error of combined parameter BP-ANN and ultrasonic parametric network were less than those of clinical parameter BP-ANN (P＜0.05), but there was no significant difference between them (P＞0.05). Conclusions：BP-ANN is better than traditional prediction method for EFW. The best model for EFW is combined parameter BP-ANN.

Key words:  Term birth, Fetal weight, Models, theoretical, Neural networks （computer）, Back propagation artificial neural network