基于Stacking集成策略的阿尔茨海默病诊断模型研究  

Multi-classification diagnosis of Alzheimer’s disease based on Stacking ensemble strategy

作　　者：韩红娟 陈杜荣[1] 秦瑶 张荣 白文琳 崔靖 马艺菲 刘龙 余红梅 HAN Hong-juan;CHEN Du-rong;QIN Yao;ZHANG Rong;BAI Wen-lin;CUI Jing;MA Yi-fei;LIU Long;YU Hong-mei(不详;Department of Health Statistics,School of Public Health,Shanxi Medical University,Taiyuan,Shanxi 030001,China)

机构地区：[1]山西医科大学公共卫生学院卫生统计学教研室,山西太原030001 [2]山西医科大学基础医学院数学教研室,山西太原030001 [3]重大疾病风险评估山西省重点实验室,山西太原030001

出　　处：《现代预防医学》2022年第22期4045-4051,4089,共8页Modern Preventive Medicine

基　　金：国家自然科学基金面上项目(81973154);山西省青年自然科学基金(201801D221399,201901D211330,20210302123242)。

摘　　要：目的针对阿尔茨海默病(AD)相关临床人群,包括认知正常(CN)、显著记忆障碍(SMC)、早期轻度认知障碍(EMCI)、晚期轻度认知障碍(LMCI)和AD进行多分类研究,以期实现AD计算机辅助诊断。方法基于阿尔茨海默病神经影像学计划(ADNI)数据库中2006例受试者(436例NC,261例SMC,323例EMCI,606例LMCI和380例AD),采用LASSO方法进行特征选择,SMOTE过采样方法处理类别不平衡问题,采用支持向量机、随机森林、逻辑回归和K近邻作为初级学习器,逻辑回归作为次级学习器,加权投票集成策略构建Stacking多分类诊断模型。结果较于以上四种初级学习器,本研究构建的Stacking集成模型分类效果较好,稳定性高,在NC vs非NC,SMC vs非SMC,EMCI vs非EMCI和LMCI vs AD之间分类准确率、召回率、F1 Score均值均在92%以上,AUC均值均在0.97以上。结论本研究构建的AD多分类Stacking集成策略,具有较好的分类性能,可科学指导AD的预防与控制,为临床医生提供自动化的AD临床辅助诊断。Objective To construct an Alzheimer’s disease(AD)auxiliary diagnosis model to classify normal cognition(NC),significant memory concern(SMC),early mild cognitive impairment(EMCI),late mild cognitive impairment(LMCI),and AD.Methods The data we used were from the Alzheimer’s Disease Neuroimaging Initiative(ADNI)with a total of 2006participants(436 NC,261 SMC,323 EMCI,606 LMCI,and 380 AD).LASSO was utilized to screen subsets of features,and SMOTE oversampling was used to address class imbalance problem of data.We built a multi-classification model using the Stacking ensemble strategy in which the support vector machine,random forest,logistic regression,and K-nearest neighbor were used as base learners and logistic regression were used as meta learners.Results Compared with four base classifiers,the Stacking ensemble model constructed in this study had the best classification effect and showed stable model performance.The mean of accuracy,recall,and F1 Score of NC vs non-NC,SMC vs non-SMC,EMCI vs non-EMCI,and LMCI vs AD were all above 92%,and the mean of AUC values were above 0.97.Conclusion The Stacking ensemble strategy of multi-classification has good classification performance,which can guide the prevention and control of AD scientifically and provide physicians with automatic clinical diagnosis of AD.

关 键 词：阿尔茨海默病 显著记忆障碍 轻度认知障碍 多分类 

分 类 号：R749.16[医药卫生—神经病学与精神病学]