Elsevier

Magnetic Resonance Imaging

Volume 22, Issue 9, November 2004, Pages 1181-1191
Magnetic Resonance Imaging

A method for comparing group fMRI data using independent component analysis: application to visual, motor and visuomotor tasks

https://doi.org/10.1016/j.mri.2004.09.004Get rights and content

Abstract

Independent component analysis (ICA) is an approach for decomposing fMRI data into spatially independent maps and time courses. We have recently proposed a method for ICA of multisubject data; in the current paper, an extension is proposed for allowing ICA group comparisons. This method is applied to data from experiments designed to stimulate visual cortex, motor cortex or both visual and motor cortices. Several intergroup and intragroup metrics are proposed for assessing the utility of the components for comparisons of group ICA data. The proposed method may prove to be useful in answering questions requiring multigroup comparisons when a flexible modeling approach is desired.

Introduction

Independent component analysis (ICA) is a method that has shown promise in characterizing the complexities inherent in fMRI data and may be especially useful for the analysis of rich naturalistic behaviors (such as simulated driving) [1]. Independent component analysis attempts to separate independent “sources” that have been mixed together (e.g., separating the voices from different speakers recorded on several microphones) using a measure of statistical independence. Independent component analysis has been successfully applied to extract components of interest from fMRI data [2], [3], [4]. The assumptions implicit in ICA have been explored and generative models of fMRI data have been proposed in order to better understand the properties of ICA as applied to fMRI data [5], [6], [7]. The primary assumption used in ICA is that of independence in space (nonsystematically overlapping networks) or time (nonsystematically varying networks) [8]. Independent component analysis can be used in a pluralistic manner along with a hypothesis-driven approach and has been successfully utilized to extract components that, although somewhat task-related, are not detected with standard general linear model (GLM) approaches [9].

In many fMRI experiments, it is desirable to directly compare and contrast different conditions either within or between subjects. Methods for performing such comparisons have been built on the GLM [10], [11], [12]. However, such approaches do not naturally extend to ICA, in which both the activation images and the time courses are estimated from the data. We have recently proposed a method for group ICA in which a single ICA group estimation is performed and individual ICA maps and time courses are then computed [13]. We proposed a straightforward extension that compared spatial or temporal features extracted from the group data. Put another way, a recent paper by Bartels and Zeki [14] states that spatial ICA reveals “chronoarchitectonically identified areas”, which, once identified, can be interrogated to compare the timings of brain activity between groups. The goal is to answer questions about the similarity or difference of components extracted from data reporting on different subjects or paradigms. We demonstrate its application to fMRI experiments designed to stimulate (a) visual (V) cortex only, (b) motor (M) cortex only or (c) visual and motor [visuomotor (VM)] cortices. We also discuss methods for assessing the utility of the components for comparison.

Section snippets

Comparing group ICA results

A general outline for a method of comparing ICA results is shown in Fig. 1. For clarity, we assume that the number of subjects within a group is the same and the same number of components is estimated for each group, although this is not a requirement of our approach. Group ICA estimation [13] is performed on two groups containing M subjects in which N components are estimated for each group. Following back-reconstruction, this produces a set of MN components for each group (N components per

Participants and paradigm

The Johns Hopkins Institutional Review Board approved the protocol and all participants provided written informed consent. Eight right-handed subjects with normal vision (7 men, 1 woman; average age, 24) participated in the study. Stimuli were projected via an LCD projector onto a rear-projection screen subtending approximately 25° of V field, visible via a mirror attached to the MRI head coil. The V paradigm consisted of an 8 Hz reversing checkerboard pattern presented for 15 s in the right V

Results

The results from the group ICA (selected component maps and time courses) for each of the paradigms are presented in Fig. 2. The S.D. of the time courses across the eight subjects is indicated with dotted lines. As expected, the V paradigm reveals activation in the left [red (R)] or right [blue (B)] V cortex during right or left V hemifield stimulation, respectively, with time courses closely matching the paradigm. The M paradigm reveals activation in the left (R) or right (B) M cortex during

Discussion

We have presented a general approach for performing statistical comparisons of ICA estimations of two or more sets of multisubject data (e.g., intergroup or interparadigm comparisons). Activation maps and their respective time courses were compared between data from three paradigms collected from eight subjects. V, M and VM paradigms were chosen to validate the ICA method. The V and M cortex activation curves were found to be highly correlated, even across paradigms. For the VM paradigm, the

Conclusion

We extend our group ICA method to enable comparisons of multisubject fMRI data and demonstrated its application to a set of V, M and VM paradigms. An approach is proposed for performing subtractive and conjunctive analyses of brain activation maps and time courses across groups or paradigms based on the flexible modeling approach, ICA. Metrics for assessing the comparison are also discussed. The proposed method may prove useful for answering questions requiring multigroup comparisons when a

Acknowledgments

Data were acquired at the F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute. Supported by the National Institutes of Health under grants 1 R01 EB 000840 (to VC) and P41 RR15241.

References (28)

  • M.J. McKeown et al.

    Analysis of fMRI data by blind separation into independent spatial components

    Hum. Brain Mapp.

    (1998)
  • M.J. McKeown et al.

    Independent component analysis of fMRI data: examining the assumptions

    Hum. Brain Mapp.

    (1998)
  • V.D. Calhoun et al.

    Independent components analysis applied to fMRI data: a generative model for validating results

    J. VLSI Signal Process. Syst.

    (2004)
  • F. Esposito et al.

    Spatial independent component analysis of functional MRI time-series: to what extent do results depend on the algorithm used?

    Hum. Brain Mapp.

    (2002)
  • Cited by (147)

    View all citing articles on Scopus
    View full text