Peter Erhard wrote:
> Hi all,recently we had a question on this mailbase, that has not been answered
> yet:
> We wish to analyse data parametrically. We acquired five fMRI runs per
> subject (8 subjects).
> Task level was varied from run to run. We are interested in the main
> task effect as well as in
> correlation of the BOLD signal with task level.
> So far we analysed our data by building one single long fMRI run by
> chaining together
> all five runs/subject.
> Then we applied a set of three orthogonalized basic functions
> (boxcar/first and second order).
> For each of this covariates a contrast image was calculated on the first
> level.
> For group analysis, we performed a one sample t-test over the
> corresponding contrast images of all 8 subjects.
>
> Our quesions are:
> 1) Is this a valid approach?
Yes.
But just to confirm what I think you're doing here's a summary
of the necessary steps.
For each subject you have 5 runs, each of which corresponded to
a particular task-level. This is your parametric variable.
For each subject you then use a contrast c=[0 0 0 0 1 0 0 ...], where
the '1' picks out the regression coefficient corresponding to the
task-level. This gives you a con*.img for each subject. You then
put the 8 con*.imgs (one from each subject) into a 'one-sample t-test'.
This then tests the hypothesis that there is a significant effect of
task-level on the BOLD activation (at each voxel) in the population
your subjects are from.
>
> 2) Would it be any better to perform parametric analysis by calculating
> contrast images for each
> run (=level)/subject on the first level and then calculating a 'simple
> (linear) regression' over the
> five level-subject-specific contrast images (total 40) on the second
> level?
No.
>
> 3) What is the appropriate analysis for the catagorical effect
> (irrespective of task level modulation)
> when applying approach 2 ?
> (one contrast image including all 5 levels/subject on the first level
Yes.
>
> or five contrast images/subject entered into one (big) t-test)
No.
>
>
You should create a single con*.img from each subject. The contrast to be
used should pick out the categorical effect in each subject eg.
c=[1 -1 0 0 .....] to pick out differences between conditions 1 and 2 (first
two columns).
>
> Hope my question is not too confusing.
>
> TIA
> Peter
All the best,
Will.
--
William D. Penny
Wellcome Department of Cognitive Neurology
University College London
12 Queen Square
London WC1N 3BG
Tel: 020 7833 7478
FAX: 020 7813 1420
Email: [log in to unmask]
URL: http://www.fil.ion.ucl.ac.uk/~wpenny/
|