Hi,
On 26 Jul 2007, at 18:31, Ethan J wrote:
> 1. I have a single group of 9 subjects and I am
> interested in testing for correlations of FA with fMRI
> signal values. From what I have learnt from the
> archives and the randomise manual, I would need to
> define a design with 2 EVs - mean for the group i.e. 1
> for all subjects and another EV for fMRI signal -
> which is demeaned. By demeaning, you basically
> calculate the average value over the group and to get
> the demeaned values I would subtract this avg value
> from the individual values, am I correct so far?
Correct. Though in randomise you could alternatively leave out the
group mean EV and tell randomise to demean the data with the -D option.
> 2. I also want to correct for age of the subjects.
> This would be added as a separate confound matrix
> rather than including in the design matrix right? So
> this confound matrix would look something like - 2EVs
> - one for the group mean and second EV containing age
> values for each subject. I believe age needs to be
> demeaned too? If I have more than one confound
> variables, would I include all of them in a single
> confound matrix or separate matrices are required?
Right - put all the confounds in a single confounds matrix (though in
the new version to be released soon in FSL 4.0 you will just include
confounds in the main design matrix, and ignore them via the contrasts).
> 3. From the archives, it looks like its recommended to
> use the maxc_tstat output to make all the
> interpretations. Randomise however produces max and
> vox and raw tstat outputs. Is it valid to use these
> outputs to interpret correlations?
Only maxc (corrected cluster-based p-values) and max (corrected voxel
p-values) are corrected for multiple comparisons across voxels, so
you would normally only use one of these - however, max is rarely
significant (because of the multiple comparison correction) so you
generally end up using just maxc.
> 4. Say EV1 is group mean and EV2 is fMRI signal, to
> study positive correlations, I am using [0 1] contrast
> and to study negative correlations I have the [0 -1]
> contrast. Right?
Right.
> 5. This question might be more related when once I
> have correctly finished these analyses. How would I
> export the TBSS clusters as ROIs? Is there a tool for
> that? Sorry I was able to find mention of this in the
> archive, but did not find the tool for doing this.
If you want to turn a maxc image into ROI clusters you can feed it
into the "cluster" program with various options allowing you then to
select individual clusters (e.g. via the -o option and then use
avwmaths++ to threshold out the nonintersting clusters).
Cheers, Steve.
------------------------------------------------------------------------
---
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
------------------------------------------------------------------------
---
|