The present invention discloses a group tensor method for multi subject fMRI data analysis, belonging to the field of fMRI data analysis. Characterized by minimizing the differences between subjects as the principle, the original construction of a tensor of all subjects fMRI data in the subjects were divided into multiple sub groups, the time course of each component were tested in each sub group between space and brain activation between components with the largest correlation. Compared with the original CP model is decomposed by large tensor, group fMRI data to construct the sub group with more matching tensor tensor, so it can be decomposed to improving the performance of multi subjects shared time process components and brain activation areas such as spatial composition, correlation coefficient and composition prior task related time stimulation process can improve the process about 0.1, noise space task related activation area of brain prime components can be reduced by about 23%, expected numbers of active voxels is almost unchanged. The invention has reference function for solving the mismatching problem of other types of high dimensional data and CP model.
【技术实现步骤摘要】
一种用于多被试fMRI数据分析的分组张量方法
本专利技术涉及一种多被试fMRI数据的分析方法,特别是涉及一种多被试fMRI数据分析的张量分解方法。
技术介绍
功能磁共振成像(functionalmagneticresonanceimaging,fMRI)被称为观察大脑的有效窗口,因为fMRI技术能够采集到被试在完成某种特定任务(task,如视觉、听觉、运动等)时的脑功能数据,且具有无损伤和高空间分辨率优势。通过采用盲源分离(blindsourceseparation,BSS)、独立成分分析(independentcomponentanalysis,ICA)等数据驱动的分析方法,无需任何先验信息,就能从fMRI数据中估计出特定任务下的多个(通常几十个)脑空间激活区(spatialactivations)成分及其时间过程(timecourses)成分,为脑功能分析和临床诊断提供详实依据。fMRI数据是高维数据。其中,单被试fMRI数据为4维,如165×53×63×46,包括3维全脑数据(53×63×46)和1维扫描次数(165);多被试fMRI数据为5维,如165×5 ...
【技术保护点】
一种用于多被试fMRI数据分析的分组张量方法,其特征是,将M个被试fMRI数据构建的一个大张量
【技术特征摘要】
1.一种用于多被试fMRI数据分析的分组张量方法,其特征是,将M个被试fMRI数据构建的一个大张量以被试为单位分成K个子组每个子组包含Nk个被试,2≤Nk<M;K个子组的被试间互相关系数平均值之和该平均值之和在M种分组方案中为最大值,为子组k内各被试的任务相关时间过程成分及脑空间激活区成分之间的互相关系数平均值,k=1,…,K;对各子组张量分别进行CP分解:得到各子组被试间共享的时间过程成分aj(k)和脑空间激活区成分sj(k)、被试间的强度差异cj(k)及残差ε(k);J为成分的个数;互相关系数计算的对象是M个被试fMRI数据的任务相关时间过程成分及脑空间激活区成分由ICA方法分离得到;求取和的协方差阵Ra和Rs,得到不同被试p、q,p,q∈{1,2,…M}时间过...
【专利技术属性】
技术研发人员:林秋华,邝利丹,龚晓峰,丛丰裕,
申请(专利权)人:大连理工大学,
类型:发明
国别省市:辽宁,21
还没有人留言评论。发表了对其他浏览者有用的留言会获得科技券。