【作者】 田敬霞；

【导师】 陈子江；

【作者基本信息】 山东大学 ， 妇产科学， 2008， 博士

【摘要】 目的:探讨人工神经网络在预测新生儿出生体重及孕妇分娩方式中的价值。方法:1.胎儿体重预测将我院226例足月、单胎、无妊娠合并症及并发症的初产妇分为训练组(100例,其中男女胎儿各50例)和验证组(126例,其中男女胎儿各63例),用训练组分别选取不同的参数构建了5个神经网络。①联合参数法,用孕妇的身高、体重、腹围、宫高及B超下胎儿的双顶径、股骨长和羊水池最大深度作为输入节点。②孕妇参数法,用孕妇的身高、体重、腹围和宫高作为输入节点。③胎儿参数法,用B超下胎儿的双顶径、股骨长和羊水池最大深度作为输入节点。④男婴联合参数法。⑤女婴联合参数法。后两组输入节点同组①联合参数法,数据则分别来自50例男婴或50例女婴。神经网络构建完成后以126例验证组来分别测试5种网络的准确性和误差,其中第④、⑤组分别用63例男婴或63例女婴测试。2.分娩方式预测研究对象全部病例来自2007年在济南中心医院分娩的足月、单胎、无妊娠合并症及并发症的初产妇,共220例样本。其中顺产58例、会阴侧切术56例、胎头吸引术48例、剖宫产58例。排除胎儿窘迫等突发因素,因为头位难产而行的手术。剖宫产者须经充分试产、宫口开全而试产失败者。将全部样本按奇、偶数随机分成两组:训练组104例,验证组116例。选取了11个输入参数,包括孕妇参数:孕妇的身高、体重、宫高、腹围,骨盆外测量的四个经线:髂棘间径、髂嵴间径、骶耻外径、出口横径。超声参数:胎儿双顶径、股骨长、羊水池深度。将训练组104例分别用一项数值法和四项分类法构建网络。结果:1.胎儿体重预测①前三组比较,联合参数法准确率最高为84.94%,母亲参数法为83.45%。胎儿参数法为80.80%。绝对误差相比,P＜0.01。②在联合参数法中,宫高的影响系数最高,为28.6%;其次为孕妇身高,为27.6%;第三位胎儿股骨长,为23.3%。羊水池深度的影响系数为8.1%。③男婴参数法与女婴参数法相比,男婴参数法为89.07%,明显高于女婴参数法的80.84%,也高于不分男女的联合参数法。方差分析两种网络的预测误差,差异有显著性(P＜0.01)。2.分娩方式预测一项数值预测法的总误差率为38.33%,四项分类预测法的总误差率为33.34%,结果相近。四项分类法非剖宫产的预测正确率高达81.18%,而剖宫产预测正确率则仅为19.35%。结论:采用人工神经网络预测胎儿体重及分娩方式有很好的研究价值和应用前景。选取合适的孕妇及胎儿参数建立网络,依男女性别分别建立网络可以提高预测的准确性。羊水量对胎儿体重预测的准确性有一定影响。更多还原

【Abstract】 Objective:To investigate the effect of Artificial Neural Networks in predicting fetal weight and labor mode.Methods:1.Fetal weight predictionThe 226 cases of full-term singleton pregnancy without complication are divided into two groups,100 samples(female and male fetuses each 50) for training and 126 samples(female and male fetuses each 63) for testing.Three groups of neural networks are composed:(1) Joint Parameter Method using the height,weight, abdominal circumference and uterine fundal height of pregnant women as well as the biparietal diameter(BPD),femur length(FL) and amniotic fluid pool depth(AFD) of fetuses under B-ultrasonography;(2) Maternal Parameter Method using the height, weight,abdominal circumference and uterine fundal height of pregnant women;(3) Fetal Parameter Method using the BPD,FL and AFD of fetuses;The neural networks are then trained and tested upon the above samples.2.Labor mode predictionAll investigating 220 cases are collected out of primiparas in full-term singleton pregnancy without complication from Jinan Central Hospital.Among them,58 are spontaneous labor,56 lateral incision of perineum,48 pull method of fetal head,and 58 caesarean sections.The caesarean section cases are regardless of cephalic presentation dystocia and should be those who had tried enough.All samples are divided randomly into two groups:104 for training and 116 for testing.Eleven parameters are chosen:8 maternal ones of the height,weight,abdominal circumference and uterine fundal height of pregnant women as well as the biparietal diameter(BPD),femur length(FL) and amniotic fluid pool depth(AFD) of fetuses under B-ultrasonography;3 ultrasonic ones of the BPD,FL and AFD of fetuses.One Output Value method and Four Output Classification method are designed to create the neural networks. Results:1.Fetal weight prediction(1) Among the three groups,the Joint Parameter Method has the highest predicting accuracy of 84.94%,the Maternal Parameter Method 83.45%and the fetal parameter method 80.80%.Variance P is less than 0.01.(2) Among Joint Parameter Method,uterine fundal heiglit has the highest impact factor of 28.6%.The second comes the height of pregnant women with 27.6%.The third femur length(FL) 23.3%and amniotic fluid pool depth(AFD) 8.1%.(3) Boy Parameter method produces a better prediction precision of 89.07 than Girl Parameter method of 80.84%,and at the same time better than the mixing method.Variance analysis illustrates significance with P＜0.01.2.Labor mode predictionOne Output Value method reaches a total error rate of 38.33%and Four Output Classification method with a favorable 33.34%.Four Output Classification method gives an accuracy of 81.18%for vaginal birth and 19.35%for caesarean section.Conclusions:The fetal weight estimation and labor mode prediction using artificial neural networks show potential research value and application prospect.A subtle combination of maternal and fetal parameters is critical to build up an effective network.Sex-specific research can produce better prediction.Amnion may reduce the prediction precision.更多还原