神经网络在预测胎儿体重中的作用

对胎儿超声参数进行神经网络处理，建立预测出生时的胎儿体重的前馈神经网络模型，方法本网由输入层，隐藏层和输出层三部分组成，输入层有９个神经元（Ｎｅｕｒｏｎｓ，Ｎｉｊ表示第ｉ怪第ｊ个神经元），分别表示ＢＰＤ，ＡＤ和ＦＬ的１０ｔｈ％，５０ｔｈ％和９０ｔｈ％。隐藏层分两层，第一层有１８个神经元，第二层有６个神经元，这两层对输入信号进行处理，输出层有３个神经元，分别代表胎儿体重的１０ｔｇｈ％、５０ｔｈ％和９

The Value of a Neural Network for the Ultrasonographic Expectation of Fetal Weight at Delivery

Objective To evaluate the ultrasonographic expection of fetal weight at delivery with a nural network computer model.Methods A four layer backpropagation neural network(BP) was built with 9 neurons in input layer,18 neurons in first hiden layer,6 neuron in second hiden layer,and 3 neurons in output layer.The Sigmoid function was used to translate code between decimol and binery number in preoperative and postoperative proccession,and the BP was trained with 21 idealised patterns.The rasults of this predictive BP model were compared with the actural fetal weights at delivery in 371 cases third tri term pregnancy fetus.Results There were significant relationship between the BP model predictive value and actual fetal weights( r =0 6850, P <0 01),the average error of prediction was 318 g(10 97%),the error in 21 2% cases were less than 100g,the error in 39 0% cases were less than 200g,in 53 6% less than 300g,in 79 26% cases less than 500g.the precision of this BP model was affected by fetal weight( P <0 05),but were not been done by gestational age,and weeks to delivery.Conclusion this BP neural network was very effecive and perspective,though the constructure and prameters need to be improved.This BP model provided a new method to treat the Ultrasonography and Obstetrics complex data.