摘要
目的探讨BP神经网络预测妊娠期糖尿病(GDM)胎儿出生体重的价值。方法将306例足月、单胎、无妊娠其它合并症及并发症的GDM孕妇随机分为训练组(200例,男女胎儿分别为106例、94例)和验证组(106例,男女胎儿分别为56例、50例)。训练组分别选取不同参数构建3个神经网络:(1)孕妇参数法:包括孕妇体重指数(BMI)、腹围、宫高、孕期增加体重、空腹血糖(FBS)、餐后2 h血糖(PBS)、糖化血红蛋白(GHbA1c)等7项参数作为输入节点;(2)胎儿参数法:用胎儿的双顶径(BPD)、股骨长度(FL)、头围(HC)、腹围(AC)、腹径(AD)、股骨皮下脂肪厚度(FTSTT)、胎儿腹壁脂肪层厚度(FFL)等7项参数作为输入节点;(3)联合参数法:将孕妇及胎儿的参数作为输入节点。神经网络构建完成后以106例验证组来分别测试3种网络法的误差率和符合率。结果联合参数法准确率最高为86.20%,胎儿参数法为71.30%,孕妇参数法为64.50%。结论BP神经网络预测胎儿体重有很好的应用前景。选取合适的孕妇及胎儿参数建立网络可提高预测的准确性。
Objective To investigate the value of BP neural network in predicting fetal birth weight in patients with gestational diabetes mellitus （GDM）. Methods 306 pregnant women of full-term pregnancy, single gestation, with no other complications of pregnancy and complications of GDM, were randomly divided into training group （200 cases, including 106 male fetuses and 94 female fetuses ） and test group （106 cases, including 56 male fetuses and 50 female fetuses）. Training group were selected to build three different neural networks with different parameters, （1） Pregnant women parameter method： including body mass index （BMI）, abdominal circumference, fundal height, pregnancy weight gain, fasting blood sugar （FBS）, postprandial blood sugar （ PBS）, glycosylated hemoglobin （ GHbA1 c）, these seven parameters were used as input nodes. （2） Fetal parameter method： including fetal biparietal diameter （BPD）, femur length （ FL）, head circumference （ HC ）, abdominal circumference （ AC ）, abdominal diameter （AD） , femoral thigh soft tissue thickness （FISTT） , and fetal abdominal wall fat layer（FFL）, these seven parameters were used as input nodes. （3） Joint parameter method ： the above maternal and fetal parameters were used as input nodes. After establishment of neural networks, the data of 106 cases of test group was used to verified prediction error rate and prediction coincidence rate of the three neural networks in predicting fetal birth weight.
出处
《广东药学院学报》
CAS
2009年第5期530-533,共4页
Academic Journal of Guangdong College of Pharmacy
关键词
妊娠期糖尿病
胎儿体重
预测
BP神经网络
gestational diabetes mellitus
fetal birth weight
forecast
BP neural network
