2型糖尿病并发肾病中西医多模态特征融合预测模型构建  

Construction of predictive model of multimodal feature fusion of traditional Chinese and Western medicine in type 2 diabetes complicated with kidney disease

作　　者：夏庭伟 李炜弘[1] 丁维俊[1] 成词松[1] 汤朝晖[1] 许强[2] 刘桠[3] 杨越 王杰鑫 XIA Ting-wei;LI Wei-hong;DING Wei-jun;CHENG Ci-song;TANG Zhao-hui;XU Qiang;LIU Ya;YANG Yue;WANG Jie-xin(School of Basic Medicine,Chengdu University of Traditional Chinese Medicine,Chengdu 610075,China;School of Intelligent Medicine,Chengdu University of Traditional Chinese Medicine,Chengdu 610075,China;Hospital of Chengdu University of TCM,Chengdu 610072,China;Sichuan Province Orthopedic Hospital,Chengdu 610041,China)

机构地区：[1]成都中医药大学基础医学院,成都610075 [2]成都中医药大学智能医学学院,成都610075 [3]成都中医药大学附属医院,成都610072 [4]四川省骨科医院,成都610041

出　　处：《中华中医药杂志》2022年第7期4116-4120,共5页China Journal of Traditional Chinese Medicine and Pharmacy

基　　金：国家科技部国家重点研发计划“中医药现代化研究重点专项”课题(No.2017YFC1703304);国家自然科学基金项目(No.81873204);中国博士后科学基金面上项目“地区专项支持计划”(No.2021M693790);成都中医药大学“杏林学者”学科人才科研提升计划(No.BSH2020013)。

摘　　要：目的:融合中西医多模态特征,构建2型糖尿病并发肾病混合深度神经网络预测模型。方法:纳入2型糖尿病无肾病患者622例、有肾病246例。采集中医四诊信息、辅助检查指标,采集舌图像。纳入患者一般信息、辅助检查指标,采用多种机器学习算法构建模型一。融合证型数据构建模型二,采用深度学习算法融合舌图像数据构建模型三。结果:采用主成分分析对辅助检查数据降维,提取20个公共因子。模型一,人工神经网络准确度、灵敏度、特异度均高于其他算法,分别为81.16%、82.57%、84.80%。模型二,人工神经网络为最优,准确度、灵敏度、特异度分别为85.13%、83.07%、85.25%。模型三,准确度88.46%,灵敏度79.36%,特异度91.51%。结论:融合中西医多模态特征的疾病预测方法,性能更佳。Objective:To integrate the multi-modal features of Chinese and Western medicine to construct a mixed deep neural network prediction model for type 2 diabetes complicated with nephropathy.Methods:Enrolled 622 patients with type 2 diabetes without nephropathy and 246 patients with nephropathy.Collect information of four Chinese medicine diagnosis,auxiliary examination indicators,and collect tongue images.Incorporate general patient information and auxiliary examination indicators,and use a variety of machine learning algorithms to build model 1.Fusion syndrome data is used to build model 2,and deep learning algorithm is used to fuse tongue image data to build model 3.Results:Principal component analysis was used to reduce the dimension of auxiliary examination data,and 20 common factors were extracted.In model 1,the accuracy,sensitivity,and specificity of artificial neural network are higher than other algorithms,which are 81.16%,82.57%,and 84.80%respectively.Model 2,the artificial neural network is the best,the accuracy,sensitivity,and specificity are 85.13%,83.07%,and 85.25%respectively.Model 3 has an accuracy of 88.46%,a sensitivity of 79.36%,and a specificity of 91.51%.Conclusion:The disease prediction method that integrates the multi-modal features of Chinese and Western medicine has better performance.

关 键 词：2型糖尿病 肾病 中西医 多模态 机器学习 

分 类 号：R587.2[医药卫生—内分泌] R692.9[医药卫生—内科学]