【摘要】 由于胎儿体重是反映胎儿生长发育情况、宫内异常妊娠情况的重要指标,因此,胎儿的估重是医生对产妇进行临床处理的一个重要依据。传统胎儿体重预测模型的构建依赖于医学知识与生理参数选择,因此构建过程不易进行复制与推广。针对这些问题,提出一种使用深度神经网络来构建胎儿体重预测模型的方法,同时介绍了从电子病历中提取相关参数的过程,以及针对数据缺失值的补全策略。实验表明,基于深度神经网络的胎儿体重预测模型优于公式预测方法与基于传统人工神经网络的模型,且提出的缺失值补全策略能够强化模型的训练,进而提高预测的准确度。最后,基于深度神经网络的胎儿体重预测模型有很强的泛化能力与通用性,为不同地区、不同医院建立个性化的预测模型提供了可行方法。更多还原

【Abstract】 Fetal weight is an important indicator which reflects the fetus’ s growth and development status,so the estimation of fetal weight becomes a crucial foundation in obstetrical decision.Most traditional fetal weight prediction models are based on medical knowledge and feature selection,which are leading to the hard repetition and promotion of the model building process.For these problems,we proposed a deep neural network structure for building fetal weight prediction model,and introduced the process in which parameters are extracted from electronic health records and the filling strategies for missing values.The experimental results show the deep neural network based prediction model outperforms traditional methods,and the filling strategy can reinforce the training of the model and improve the accuracy.Finally,the generalization ability and universality of the deep neural network model can help different areas and hospitals to build personalized fetal weight prediction model.更多还原