I'm not really qualified to answer that, but basically the change would have to be *really* big or your design *really* imbalanced. The DMN would have to be so radically different/underrepresented as to not generate a component at all. As long as can get a DMN component, I think you are good to go for comparisons. And since the DMN is so robust, you shouldn't really have to worry about all this.
On the flipside, doing one big group ICA + dual regression does have an advantage. All of your connectivity maps will be based on the same component and have the same scale. So if you do end up with four separate fixed-effect tests instead of two mixed-effects tests, you will still be able to meaningfully compare regression coefficients and t-values across tests.
On Sunday, July 15, 2012 17:50:10 you wrote:
> Hi Ben,
>
>
> Thanks for the prompt reply. However, I do have one concern....my hypothesis is that there may be a change in the DMN, if I combine all of the pre post experimental and control together to run a ICA, will it affect my results? let's say if there is indeed a change, will this cause it to disappear or become less significant?
>
> Thanks,
> Catherine
>
>
>
> ________________________________
> From: Benjamin Kay <[log in to unmask]>
> To: [log in to unmask]
> Sent: Sunday, July 15, 2012 2:37 PM
> Subject: Re: [FSL] Question regarding Concat-ICA and dual regression
>
> Opinions may differ on this, but I would be inclined to concatenate all your inputs, (pre, post, experimental, and control) into one ICA + dual regression analysis. Once you've identified your DMN component and have a DMN connectivity map (backprojected using dual regression) for each input, you can apply whatever statistical model you like to test your hypotheses. For example, for pre- v.s. post-intervention, you could do a paired t-test with group as a random factor, or you could do two separate paired t-tests, one for each group. Doing the latter does not obligate you to run a separate ICA analysis for each group.
>
> On Sunday, July 15, 2012 17:24:57 you wrote:
> > Dear All,
> >
> > I'm new to the ICA analysis, and would like to ask for some advice regarding how to approach the analysis...
> > I have two groups(1 experiment,1 control, all are healthy students), each has a pre and post scan before and after the intervention. I want to test if there is a change in the DMN after the intervention both within each group and between groups.
> >
> > Here are the steps I'm planning to do, wandering if they are correct:
> > 1.test if there is a change before and after the intervention for each group
> > i.e. run concat-ica for all pre & post conditions of the experiment group together(15 subjects, 30 inputs) and in the GLM setup, create two EVs (pre, post) and do contrasts?
> > OR
> > run concat-ica for pre and post conditions separately and do contrasts later?
> >
> > As for between groups, should I do concat-ica with dual regression instead?
> > i.e. run concat-ica for all post or pre conditions (experiment+control) to create a template, and then estimate subject-wise, etc....
> >
> > Thanks,
> > Catherine
|