SCIENCE@home

Spatial ICA reveals functional activity hidden from traditional fMRI GLM-based analyses

Jiansong Xu,M. Potenza,V. Calhoun

Published 2013 in Frontiers in Neuroscience

ABSTRACT

Independent component analysis (ICA) isasignalprocessingtechniqueusinghigher-orderstatisticstoextractsignalsbyunmix-ingsignalmixtures.McKeownetal.(1998)introduced spatial ICA (sICA) into func-tionalmagneticresonanceimaging(fMRI)study in the late 1990s. SICA assumes thatfMRI signal from each voxel represents alinear mixture of source signals, separatesthis signal mixture into spatially inde-pendent source signals using higher-orderstatistics, and groups all brain regionsshowing synchronized source signals intoindependent components (ICs), whichrepresent temporally coherent functionalnetworks(FNs)(McKeownandSejnowski,1998;McKeownetal.,1998;Calhounetal.,2002, 2009). McKeown et al. (1998) pre-dicted that sICA would be more sensitivein detecting task-related changes in fMRIsignal than the traditional general lin-ear model (GLM) based analysis, becausesICAusesadata-drivenapproach,andcanreduce noise in the final solution by sepa-ratingartifacts from real fMRI signal.SICA hasbecome widely used for fMRIanalysis since its original application tofMRI 15 year ago (Calhoun and Adali,2012). Most studies use sICA as a tech-nique for extracting FNs from fMRI data,whilesomestudiesusesICAtoseparateand remove artifacts from real fMRI sig-nal for improving the sensitivity of sub-sequent GLM based analysis (e.g., Aronand Poldrack, 2006). Several studies com-pared sICA and GLM based analyses andreported that sICA revealed more brainregions showing task-related activationrelativetoGLMbasedanalysis,supportingthepredictionofMcKeownetal.(Calhounet al., 2001; Malinen et al., 2007; Tie et al.,2008; Kim et al., 2011). Relevant to thisissue, an interesting finding from sICAis overlap of two or more FNs of differ-ent timecourses and different task-relatedmodulations. This finding has been inter-preted as evidence of multiple concurrentcentral processes associated with commonbrain regions (Calhoun et al., 2008; Kimet al., 2009a,b; Menz et al., 2009; vanWageningenetal.,2009; Wuetal.,2009; StJacquesetal.,2011;Domagaliketal.,2012;Zhang and Li, 2012). Two recent studies,onefromus(Xuetal.,2013) andtheotherfrom Beldzik et al. (2013) systematicallyanalyzed FN overlap, and the task-relatedmodulation of the timecourses of over-lapping FNs. These analyses further high-lighted the presence of brain functionalactivity hidden from the traditional GLMbased analysis.In Xu et al. (2013),sICAwasusedto extract FNs from fMRI data relatedto a visual target-identifying task. Most(∼78%) functional brain regions showedoverlap of two or more FNs. Several brainregions including the dorsal anterior cin-gulate (dACC), insula and adjacent ven-trolateral prefrontal cortex (PFC), lateraltemporal and parietal cortex, and pre-cuneus/posterior cingulate (PCC) showedoverlap of seven or more FNs. Each ofoverlapping FNs showed unique task-related modulations of timecourses, andsomeofthemwere opposite(i.e.,increasesvs.decreases)toeachother.Thepredictionthat such opposite modulations withinthesamevoxelsmightbehiddenfromaGLM based analysis due to cancellationwas tested by analyzing the same fMRIdata using a GLM based analysis. Thebrain regions showing task-related activa-tion and deactivation as revealed by theGLM based analysis are 11.9 and 26.2%,respectively, of brain regions showing cor-responding changes as revealed by sICA,therefore supporting the prediction.In Beldzik et al. (2013),theauthorsused both sICA and a GLM-based analysistoassessanfMRIdatasetrelatedtoananti-saccadic task. They developed a tool calledContributive Sources Analysis (CSA) forestimating the amplitude of fMRI sig-nal changes in each FN. They first usedthe GLM-based analysis to define clustersshowing significant task-related activity asregions of interests (ROIs) and then usedCSA to extract measures of task-relatedchanges in fMRI signal within these ROIsfrom all FNs overlapping with these ROIs.They confirmed their prediction that thesum ofthese measures from all FNsequalsto the measure of task-related changes offMRI signal within the ROIs as assessedby the GLM-based analysis. They fur-ther demonstrated that task-related oppo-site modulations of overlapping FNs con-tributed to the negative findings of theGLMbasedanalysisatsome brainregions.Therefore, findings from both studies arecomplementary, and indicate that mul-tiple neural circuits, each with uniquetimecourse and task-related modulation,can occur concurrently within the same

PUBLICATION RECORD

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-41 of 41 references · Page 1 of 1

CITED BY

Showing 1-90 of 90 citing papers · Page 1 of 1