基于BP神经网络的绝经后女性骨质疏松预测模型构建  

Establishment of osteoporosis risk prediction model for postmenopausal women based on BP neural network

作　　者：朱彦霞 华南 刘小侃 ZHU Yanxia;HUA Nan;LIU Xiaokan(He'nan General Hospital,Zhengzhou 450002,He'nan Province,China;CBN Intelligent Media,China Radio and Television He'nan Network Co.,Ltd.)

机构地区：[1]河南省职工医院,郑州450002 [2]河南中广智媒科技有限公司,中国广电河南网络有限公司

出　　处：《中国数字医学》2023年第8期95-102,共8页China Digital Medicine

基　　金：2021年度河南省医学科技攻关计划联合共建项目(LHGJ20210261);2022年度河南省重点研发与推广专项(软科学)项目(222400410151、222400410184);2023年度重点研发与推广专项目(232102310491)。

摘　　要：目的:构建基于反向传播(BP)神经网络的独立于骨密度的绝经后女性骨质疏松联合预测模型,并分析其效果。方法:选取河南职工医院2020年―2021年183例包含绝经后女性骨矿含量、骨密度等检测变量的2832条测量数据,通过单因素、多因素分析方法,筛选出除骨密度之外的影响骨质疏松的独立影响因素作为特征变量,输入BP神经网络建模。结果:设置不同的训练集与测试集比例,以骨矿含量、检查范围、解剖部位、检查时的年龄、绝经年龄、身高、体质量等因素作为联合特征因子,输入神经网络模型并进行预测,其正确率均>70%,受试者工作特征曲线下面积均>0.9,与基于骨密度预测的受试者工作特征曲线下面积相当。结论:该模型选取的筛选特征因子能反映绝经后女性骨质疏松的状态,与骨密度预测效果相近,可作为预测骨质疏松的有效方案,能减少骨密度未标化可能带来的误诊。Objective To establish an osteoporosis risk prediction model independent of BMD(Bone Mineral Density)for postmenopausal women based on BP neural network and analyze its effect.Methods A total of 2,832 measurement data containing test variables such as bone mineral content and BMD in 183 cases of postmenopausal women in Henan General Hospital from 2020 to 2021 were selected,and the independent influencing factors affecting osteoporosis other than BMD were screened out as characteristic variables by single-factor and multi-factor analysis methods and input into BP neural network modeling.Results Different ratios of training and test sets were set,and factors such as bone mineral content,examination range,anatomical position,age at examination,age at menopause,height and weight were used as joint characteristic factors,which were input into the neural network model and then prediction was conducted,with accuracies>70%and AUC values>0.9,which were comparable to the AUC of BMDbased prediction.Conclusion The screening characteristic factors selected by this model can reflect the status of osteoporosis in postmenopausal women and the prediction effect is similar to that if BMD,which can be used as an effective solution for osteoporosis prediction and can reduce the possible misdiagnosis caused by unstandardized BMD.

关 键 词：神经网络 骨质疏松 骨矿含量 

分 类 号：R319[医药卫生—基础医学]