基于支持向量回归的新型胎儿体重预测模型的研究  

A new fetal weight prediction model research based on support vector regression

作　　者：刘腾[1] 朱屹 林燕茹 朱琴 唐文波 庄佳衍 LIU Teng;ZHU Qi;LIN Yanru;ZHU Qin;TANG Wenbo;ZHUANG Jiayan(The Affiliated Hospital of Medical School,Ningbo University,Zhejiang Ningbo 315020,China;Ningbo Institute of Industrial Technology,Chinese Academy of Sciences,Zhejiang Ningbo 315201,China;Ningbo University,Zhejiang Ningbo 315211,China)

机构地区：[1]宁波大学医学院附属医院,浙江宁波315020 [2]中国科学院宁波工业技术研究院,浙江宁波315201 [3]宁波大学,浙江宁波315211

出　　处：《中国妇幼健康研究》2022年第9期14-21,共8页Chinese Journal of Woman and Child Health Research

基　　金：宁波市自然科学基金资助项目(202003N4220)。

摘　　要：目的 探讨开发一种基于支持向量回归(SVR)的新型胎儿体重预测模型(简称新模型),并与经典胎儿体重预测模型对比,评估新模型在预测胎儿出生体重方面的性能。方法 收集2020年1月至6月期间在宁波大学医学院附属医院分娩的1 442例孕产妇为研究对象,在分娩前1周内采集相关数据,共计18维。通过预实验,选出最优算法(即SVR)。通过计算每个特征与新生儿出生体重的皮尔逊相关系数及SVR算法的权重,筛选建模参数。应用SVR对数据进行建模,建立基于SVR的新模型,并与经典胎儿体重预测模型公式对比,评估新模型性能。结果 分别计算每个特征与新生儿出生体重的皮尔逊相关系数,应用SVR算法的权重,结合临床经验,反复进行实验后,最后筛选得到建模所需的11维数据为:身高、体重、孕期增重、宫高、孕妇腹围、双顶径、头围、股骨长、胎儿腹围、羊水指数和采集时间(采集时间距分娩的天数)。应用SVR对11维参数进行建模,构建出新模型。与经典模型对比,预测误差在5%以内的胎儿体重预测准确率为54.48%,预测误差在10%以内的为88.28%,预测误差在250g以内的为73.10%,新模型具有最高的胎儿体重预测准确率。此外,胎儿体重预测新模型还具有最低的平均绝对百分比误差(MAPE)5.24%,平均绝对误差(MAE)178.43g和均方根误差(RMSE)230.15g。结论 新模型对胎儿体重预测的性能良好,未来可进一步在临床推广,但仍需更多的临床试验加以验证。Objective To develop a new fetal weight prediction model(the new model) based on support vector regression(SVR),and to evaluate the accuracy of the new model in predicting fetal birth weight compared with classical fetal weight prediction formulas.Methods 1442 cases of pregnant women who gave birth in the Affiliated Hospital of Medical School of Ningbo University from January 2020 to June 2020 were selected as the research objects, and relevant data were collected within 1 week before delivery, with a total of 18 dimensions.The optimal algorithm was(SVR) selected through pre-experiment.The parameters were selected by calculating the Pearson correlation coefficient between each feature and the neonatal birth weight.SVR was applied to model the data, and a new model based on SVR was established, compared with the classical fetal weight prediction model formula, to evaluate the new model performance.Results Pearson correlation coefficient of each feature and neonatal weight, weights of applied SVR algorithm combined with clinical experience, 11 parameters was selected eventually after repeated experiments.The 11-dimensional parameters were height, weight, gestational weight gain, uterine height, maternal abdominal circumference, biparietal diameter, head circumference, femur length, abdominal circumference, amniotic fluid index and collection time(the number of days between collection time and delivery).New model was established based on SVR.Compared with the classical model, the prediction accuracy of fetal birth weight within 5% was 54.48%,the prediction error within 10% was 88.28%,and the prediction error was 73.10% within 250 g, and the new model had the highest fetal weight prediction accuracy.In addition, the new model for fetal weight prediction also had the lowest mean absolute percentage error(MAPE) of 5.24%,mean absolute error(MAE) of 178.43 g and root mean square error(RMSE) of 230.15 g.Conclusion The new model has good performance in predicting fetal weight, which can be further popularized in clinical p

关 键 词：胎儿体重 出生体重 预测模型 支持向量回归 

分 类 号：R17[医药卫生—妇幼卫生保健]