基于静息态fMRI区分健康老年人认知水平的MVPA方法研究  

MVPA method study for distinguishing the cognitive level of healthy elderly people based on resting-state fMRI

作　　者：汪方毅 唐杰庆 刘倩[1,2] 余成新 李博[3] 丁帆[1,2] WANG Fangyi;TANG Jieqing;LIU Qian;YU Chengxin;LI Bo;DING Fan(Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering,China Three Gorges University,Yichang 443002,China;College of Computer and Information Technology,China Three Gorges University,Yichang 443002,China;Department of Radiology,the First College of Clinical Medical Science of China Three Gorges University and Yichang Central People's Hospital,Yichang 443003,China)

机构地区：[1]三峡大学水电工程智能视觉监测湖北省重点实验室,宜昌443002 [2]三峡大学计算机与信息学院,宜昌443002 [3]三峡大学第一临床医学院宜昌市中心人民医院放射科,宜昌443003

出　　处：《磁共振成像》2023年第6期18-25,共8页Chinese Journal of Magnetic Resonance Imaging

基　　金：湖北省教育厅科学技术研究计划(编号:Z2019096);中国高校产学研创新基金(编号:2019ITA03043)。

摘　　要：目的 基于静息态功能磁共振成像(functional magnetic resonance imaging, fMRI)数据构建脑网络,建立多变量模式分析(multivariate pattern analysis, MVPA)实现健康老年人认知水平的有效区分。材料与方法 使用公开数据集中55名认知优(认知优组)和43名认知差(认知差组)的健康老年人,基于全部受试者的静息态fMRI数据,建立MVPA方法对健康老年人认知水平进行区分。其中,脑网络构建和特征选择使用高斯Copula互信息(Gaussian Copula mutual information, GCMI),结合支持向量机(support vector machine, SVM)完成分类,然后与现有的MVPA方法区分健康老年人的认知水平作对比。通过独立样本t检验分析组间一致性功能连接的差异。结果 本文建立的MVPA方法实现了健康老年人认知水平的有效区分,分类准确率达到77.22%,敏感度、特异度及曲线下面积(area under the curve, AUC)分别为81.82%、72.09%和0.77。另外,认知差组的一致性功能连接强度相对于认知优组有明显降低,且组间差异普遍具有统计学意义(P<0.05)。结论 使用对非线性敏感的GCMI进行脑网络的构建并选择特征,结合SVM可以实现健康老年人认知水平的有效区分。通过分析一致性功能连接,发现其强度减弱可能导致认知水平降低,差异具有统计学意义的一致性功能连接对于辅助临床诊断具有重要价值。Objective:To construct brain network based on resting-state functional magnetic resonance imaging(fMRI)data and establish multivariate pattern analysis(MVPA)to effectively distinguish the cognitive level of healthy elderly people.Materials and Methods:Using a publicly available dataset of 55 healthy elderly individuals with excellent cognitive abilities(cognitive excellent group)and 43 individuals with poor cognitive abilities(cognitive poor group),based on the resting-state fMRI data of all participants,an MVPA method was established to distinguish the cognitive level of healthy elderly people.Among them,Gaussian Copula mutual information(GCMI)was used for brain network construction and feature selection,and support vector machine(SVM)was used to complete classification.Then the existing MVPA method was used to distinguish the cognitive level of healthy elderly people as a comparison.Finally,the differences of consistency functional connection between the groups were analyzed through independent-samples t-test.Results:The MVPA method established in this article effectively distinguishes the cognitive level of healthy elderly people,with a classification accuracy of 77.22%,the sensitivity,specificity and AUC were 81.82%,72.09%and 0.77 respectively.In addition,the consistency functional connection strength of the cognitive poor group was significantly reduced compared to the cognitive excellent group,and the differences between groups were generally statistically significant(P<0.05).Conclusions:Using GCMI,which is sensitive to nonlinear,to construct brain networks and filter features,combined with SVM can effectively distinguish cognitive level of healthy elderly people.Through the analysis of consistency functional connection,we found that the weakening of its strength may lead to the reduction of cognitive level.The consistency functional connection differences with statistically significant has important value in assisting clinical diagnosis.

关 键 词：功能磁共振成像 健康老年人 认知水平 分类 多变量模式分析 脑网络 高斯Copula互信息 

分 类 号：R445.2[医药卫生—影像医学与核医学] R745.1[医药卫生—诊断学]