Hi,
This is technically possible, though I think you would need to be careful at the higher-level comparison stage (stage 3 in dual regression) to account for differences, e.g. if you end up having data sets of different length then at the between-run comparison you should add a confound regressor - if you are looking for differences in Z you should be adding a column containing the sqrt(N) as a confound value, i.e. adjust for the fact that in longer runs you expect an inflation of the Z that sclaes with sqrt(Z).
Easiest at the interpretation level is to declare one group as the baseline (these might be the controls in a patient vs control study or the subjects off intervention in paired comparison) and then derive the group ICA maps from this group only. After that, you can run dual regression against this set of spatial maps for both group A and B to get subject-specific estimates which then enter in the between group stage. The difference in TR is entirely irrelevant, I guess you mean that they are of different length.
hth
Christian
On 1 Mar 2011, at 19:58, Monica Wey wrote:
> Dear all,
>
> I was wondering is it possible to combine the results of multiple resting-state fMRI data sets with different imaging parameters (e.g. TR, number of repetitions, etc.) together?
> The Biswal et al., PNAS, 2010 paper seems to be able to do so.
>
>
> If I processed two groups of data with MELODIC and got 20 components from each group.
> Any idea on how to bring these two together to get a group mean?
>
> Group A, TR=3sec, 12 subjects
> Group B, TR=2.5sec, 8 subjects
>
> Since the IC may not match up, I cannot simply use the GLM group analysis.
> And, because they have different TR, I cannot use dual regression to get the group mean.
>
> Thanks in advance for any advice.
>
> Best,
> Hsiao-Ying (Monica)
|