Hi
Yes, that's one way of doing things. You could also try and simple concatenate the design.mat files (only the design portion). You need to get rid of the header first using 'tail' and 'head' in unix and then concatenate everything into one big file
for i in subjects*; do
cat ./$i/featdir/design.mat | tail -n 180 >> all_designs.txt
done
Assuming that your experiment is of length 180vols
Then simply run fsl_glm against melodic_mix and the newly created all_design.txt to test things. This effectively gives a fixed-effects answer only (due to the inheren averaging across subjects) but saves the timecourse extraction.
For the comparison of EV1 and EV2 you would want both in the same analysis and then use contrasts (-c option in fsl_glm)
hth
Christian
On 1 Mar 2010, at 22:51, Eric Claus wrote:
> Dear FSL experts,
> I have run a multi-session temporal concatenation analysis using melodic on
> 20 subjects who completed in decision making task while undergoing fMRI.
> The analysis ran fine, but now I am trying to figure out whether it is
> possible to link each component timecourse on a subject by subject basis to
> the given subject's design matrix (from an earlier feat analysis). I am
> wondering if what I am doing to link timecourses with components makes
> sense; any feedback would be greatly appreciated.
>
> 1. For each subject, extract relevant timecourse from the report/t*.txt.
> 2. Extract convolved timecourse from design.mat file for each subject at
> the firstlevelanalysis.feat directory for each of 10 EVs.
> 3. Enter subject's timecourse as the dependent variable, and 10 EVs as
> independent variables (with no intercept term) into a multiple regression.
> 4. Extract parameter estimates for each EV for each subject.
> 5.a Compute mean parameter estimate across all subjects for a given
> component to determine if the component is sigificantly related to the EV of
> interest.
> b Compare parameter estimates across EV's (e.g. EV1 vs. EV2) for a given
> component to determine if the component is significantly more active in EV1
> or EV2.
>
> Does this approach seem reasonable? If not, is there a better approach for
> linking the components to conditions of interest, given that each subject is
> expected to have a different timings for a given EV?
>
> Thanks in advance,
> Eric
|