摘要
目的探讨轻度认知障碍(MCI)人群的目络特征,并基于此运用机器学习算法构建MCI辨识模型,为MCI早期辨识提供客观性依据。方法将2022年4—12月在福建省福州市5个社区和福建中医药大学附属第二人民医院健康管理中心招募的316例受试者作为研究对象,按照性别、年龄、受教育年限进行1∶1倾向性得分匹配,分为认知正常组和MCI组,每组158例。利用一般人口资料表、神经心理学测试量表、中医证素辨识系统和博奥目诊仪,采集研究对象的基本信息、认知测试结果、目络特征信息和中医证素。采用两独立非参数检验和卡方检验,分析MCI组和认知正常组间的目络特征和中医证素特征的差异;运用频数分析法、主成分分析法探讨MCI患者中医证素分布特点。继而将316例研究数据随机分为80%训练集和20%验证集,运用支持向量机、决策树、人工神经网络、随机森林算法,以MCI目络特征和中医证素为自变量,是否MCI为因变量,分别构建不同的MCI辨识模型,通过比较模型性能,选择最优模型,实现临床早期辨识。结果MCI组重要目络特征中,红色点、黯褐色斑、黯黄色雾漫、丘、血脉红色、血脉迂曲分布频率均高于认知正常组(P<0.05),但黯粉色斑频率低于认知正常组(P<0.05)。MCI组心、脾、肝积分均高于认知正常组(P<0.05)。MCI组的痰、湿、血瘀、气滞、热、阳亢、暑、食积积分均高于认知正常组(P<0.05)。MCI组的阴虚、血虚、气虚、津亏、气陷积分均高于认知正常组(P<0.05)。MCI组常见病位证素为肝、肾、筋骨、脾、心和经络等,常见实性证素为痰、湿、血瘀和气滞,常见虚性证素为阴虚、血虚、气虚和阳虚。基于目络特征和中医证素的MCI辨识模型,本研究选择支持向量机模型为最优判别模型,其准确率为73.08%,敏感度为0.677,特异度为0.765,AUC为0.807,整体性能较好。结论MCI患者相比于正常人有
Objective To investigate the eye collateral channel characteristics of mild cognitive impairment(MCI)population,and to build an MCI identification model based on machine learning algorithms to provide an objective basis for early recognition of MCI.Methods A total of 316 subjects from 5 communities in Fuzhou City,Fujian Province and the Health Management Center of the Second People's Hospital Affiliated to Fujian University of Traditional Chinese Medicine were recruited from April to December 2022.Eligible subjects were matched on a 1∶1 propensity score according to sex,age,and years of education and were divided into MCI group and normal cognition group,with 158 cases in each group.Using the general demographic data sheet,neuropsychological test scale,syndrome identification system of TCM and Boao visual diagnosis instrument,basic information,cognitive test results,and information on eye collateral channel characteristics and TCM symptom elements of the subjects were collected.Two independent non-parametric test and chi-square test were used to analyze the differences between the MCI group and the normal cognition group in terms of eye collateral channel characteristics and TCM symptom elements.Frequency analysis and principal component analysis were used to explore the distribution characteristics of TCM symptom elements in MCI patients.The study data of 316 cases were then randomly divided into 80%training set and 20%validation set.Different MCI identification models were constructed using support vector machine,decision tree,artificial neural network and random forest algorithm,with MCI eye collateral channel characteristics and TCM syndrome elements as independent variables and onset of MCI as a dependent variable.By comparing the model performance,the optimal model was selected to achieve early clinical recognition.Results The important eye collateral channel characteristics of the MCI group were red dots,dull brown spots,dull yellow fog diffusion,mounds,red blood veins,and tortuous blood veins(P<0.05),th
作者
吴铁成
曹蕾
尹莲花
何友泽
刘志臻
杨敏光
徐颖
吴劲松
WU Tiecheng;CAO Lei;YIN Lianhua;HE Youze;LIU Zhizhen;YANG Minguang;XU Ying;WU Jinsong(Rehabilitation Industry Institute of Fujian University of Traditional Chinese Medicine,Fuzhou,Fujian 350122,China;The Second People's Hospital Affiliated to Fujian University of Traditional Chinese Medicine,Fuzhou,Fujian 350003,China)
出处
《康复学报》
CSCD
2024年第1期76-83,共8页
Rehabilitation Medicine
基金
国家自然科学基金青年项目(82104970)
福建省中青年教师教育科研项目(JAT210213)
福建中医药大学校管课题(X2021006)
福建中医药大学青年科技创新人才培育计划(XQC2023001)。
关键词
轻度认知障碍
目络
机器学习
支持向量机
mild cognitive impairment
eyes collateral channels
machine learning
support vector machine
