Hi all,
This is perhaps getting off SPM topic (ok, it definitely is), but is an important point to make. Having two or more groups with group ICA works quite well (indeed with groups you deal with similar issues as looking for changes associated with a subject variable like IQ) and does not motivate a different back-reconstruction approach per se. It's all about capturing individual differences (i.e. if you do this well you also capture group differences).
As described in our original group ICA paper, a back-reconstruction approach (e.g. PCA-based as in GICA1-3 or regression based as in dual regression (called spatiotemporal regression in GIFT)) are needed for estimating individual subject differences in group ICA. All have similar issues wrt scaling and thresholding which are dealt with in various ways in the various packages, e.g GIFT/MELODIC/etc. All give you very similar results, you can show one or the other is slightly better or worse if you push the assumptions (STR starts with the mean spatial maps and suffers when the mean maps don't reflect small details, GICA starts with the component timecourses for each subject). Indeed GICA1 and DR/STR are mathematically identical when you have a high number of PCA dimensions at the subject level, which is what we typically recommend. For a detailed comparison of these kinds of issues please see:
E. Erhardt, S. Rachakonda, E. Bedrick, T. Adali, and V. D. Calhoun, "Comparison of multi-subject ICA methods for analysis of fMRI data," Human Brain Mapping, vol. 12, pp. 2075-2095, 2011.
And for some detailed simulations showing how group ICA captures individual differences (including group differences) see:
E. A. Allen, E. Erhardt, Y. Wei, T. Eichele, and V. D. Calhoun, "Capturing inter-subject variability with group independent component analysis of fMRI data: a simulation study," NeuroImage, vol. 59, pp. 4141-4159, 2012
Sorry for the self-citing, but hopefully this has been helpful.
Regards,
Vince
> -----Original Message-----
> From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]] On
> Behalf Of Rainer Boegle
> Sent: Tuesday, March 13, 2012 5:20 AM
> To: [log in to unmask]
> Subject: Re: [SPM] t-test on ICA results
>
> Dear Mojdeh,
>
> you may want to consider using ICA & Dual Regression as you have two groups.
>
> See this page for explanation:
> http://www.fmrib.ox.ac.uk/analysis/dualreg/
> and
> http://www.fmrib.ox.ac.uk/analysis/dualreg/dualreg_poster.pdf
>
>
> hth
> Rainer
|