功能磁共振数据的独立成分分析.pptx

Zhejiang Key Laboratory for Research in Assessment of Cognitive ImpairmentsIndependent component analysis (ICA)功能磁共振数据的独立成分分析 基础理论 方法学问题和解决 基础和临床研究应用 软件和使用张寒 研究员napoleon1982@hanzhang@QQ:utlineBasic principle ICA不是一个特别困难和难解释的方法Methodological Issue Caveat 用ICA时需要注意的问题，在发表文章时通常需要涉及Applications 如何应用ICA进行心理学和临床研究的实例(特别是术前定位)Toolkits Usage ICA软件怎么使用(上机演示) OutlineBasic principleWhat’s ICAWhat ICA producesICA on resting-state fMRI: RSNsMethodological Issue CaveatToolkits UsageApplications 鸡尾酒会的例子：ICA能重现每个音源单独发出的声音ICAJaakko S?rel?, Patrik Hoyer and Ella Bingham, graphic design by Petri Saarikko. 任务激活区检测的例子：ICA能得到和SPM/GLM类似的结果Mckeown et al., 1998,PNAS, “Stroop color-naming task”In many cases we need ICADifferent brain areas/subjects have different HRF;Special tasks: listening music, watching movie;Resting-state data;Data with heavy noise and artifacts;You are not only interested with task-evoked effect, but also other brain networks, e.g., transient effect;6. Very complicated task design相比广义线性模型(GLM): ICA的优缺点 General Linear ModelPros: a known model easy to interpretCons: unknown HRF unable if without hypothetical model ignore relationship between voxels ICAPros: without need of HRF or hypothesis count for relation between voxels find underlying data structureCons: hard to interpret instability problemsMcKoewn et. al., 1998, Hum.Brain.Mapp.Auditory EXE DMN WMPrimary motorVarious noise StriatePrimary visual静息态fMRI检测大脑功能网络的例子ICA applied to resting-state fMRI: Functional connectivity, Resting-state networks (RSNs)One of the “two most commonly used methods in functional connectivity MRI studies”. (Zhang Raichle, 2010, Nature Review)“The second most popular technique for analysing spontaneous BOLD data” with “tremendous promise”. (Fox Raichle, 2007, Nature Review Neurosci)van de Ven et al., 2004Kiviniemi et al., 2003Jafri et al., 2008Damoiseaux et al., 2006Calhoun et al., 2009Damoiseaux et al., 2008Greicius et al., 2004Calhoun et al., 2008Calhoun et al