Hello  Jeanette,

Thanks a lot for your answer.  So I don't need to run a second level, right? if so, How I can get the mean positive and negative correlation for each condition (pre and post)?  if I would want to mask the real pre> post positive/negative correlation result and post> pre positive/negative correlation result.

Just to understand and try to increase my knowledge, I would greatly  appreciate if you can explain to me, when is it necessary to do a second level analysis in fMRI analysis? when using repeat measurements??? I have read some previous papers using seed-based analysis in FSL (FEAT) they did not use a second level, but they were unpaired t test designs for instance Castellanos et al, 2008. 

Is it the analysis I have described in my prior post incorrect?

Thank you so much, I appreciate your help on this matter

Lorena 
______________________
Lorena Jiménez Castro, MD
Postdoctoral Fellow
UTHSCSA
===================
What would possibly be easiest would be to simply use fslmaths to calculate your paired differences for each subject.  Tricking Feat to use the zstats instead of copes to create the difference may be unnecessarily difficult.

To clarify, I'm assuming you have 2 first level analyses for each subject:  pre/post.  Step 1, make sure you transform your zstats from these first level analyses to standard space (easily done with flirt and files in the first level reg directory...see below).  Use fslmaths to calculate the paired differences (pre-post) of your zstats that have been transformed to standard space.  Then use fslmerge -t to create a singe 4d image (1 image per subject).  Last use the randomise call you specified.  Repeat with the difference in the opposite direction.

If you take a look in your reg directory for a single first level analysis you'll find a file called, "example_func2standard.mat".  Assuming you are in the stats directory,
flirt -in zstat#.nii.gz -ref $FSLDIR/data/standard/MNI152_T1_2mm.nii.gz -applyxfm -init ../reg/example_func2standard.mat -out zstat#_standard

will create your standard space zstat image, where you'll need to replace "#" with the correct number of the zstat you're interested in.

Cheers,
Jeanette

On Wed, May 16, 2012 at 10:26 AM, Lorena Jimenez-Castro <[log in to unmask]> wrote:
Dear Tom and Jeanette,

I would like to confirm that I understand what you wrote and commented in previous  posts. So my understanding is that if I am doing for instance a seed-based correlation analysis on a group of subjects with pre and post treatments data. I would need to run a second level analysis in FEAT including all of the fist level analysis together as explained at:
http://www.fmrib.ox.ac.uk/fslcourse/lectures/practicals/feat2/index.htm instead of doing the second level analysis for each subjects separately (as Tom described previously), using a paired t-test,  feeding the COPEs from the first level analysis.
i.e, input pre_subjects COPEs and post_subjects COPEs; design paired-t-test; output:  COPE1 pre> post for all os the subjects, COPE 2: Post > pre for all of the subjects

 I would then have to run two randomize (or FEAT) separately for each COPE (from the second level) as follow:

1) randomise  -i COPE1 pre> post for all subjects -o pre >post.nii -1 -T -m <mask>

2) randomise  -i COPE2 Post > pre_for_all subject  -o Post >pre.nii -1 -T -m <mask>


Then I would  like to mask he positive correlation and negative correlations, I think  using the contrast means from the second level

Questions:
A) Am I on the right track?

B) Following the web page http://www.fmrib.ox.ac.uk/fslcourse/lectures/practicals/feat2/index.htm, and the example described above,  Is it possible to finish the analysis in the second level and to try to publish those results?

C)  If not, Why is it necessary to perform a third level analysis in this case?

I would really appreciate your comments and advice on this matter

Thank you very much

Lorena Jiménez Castro, MD
Postdoctoral Fellow
UTHSCSA



========================

Re: [FSL] Round 2: Change in functional connectivity between pre- and post-conditions using randomise

FROM; Tom Johnstone

TO: [log in to unmask]

Message flagged Wednesday, May 16, 2012 10:05 AM
Of course you're right Jeanette. Though running the fixed effects
won't hurt either I suppose. Whichever is easiest then.

-Tom



On Wed, May 16, 2012 at 3:42 PM, Jeanette Mumford
<[log in to unmask]> wrote:
> Hi,
>
> Using fixed effects on a zstat image isn't necessary for 2 reasons.  Mostly
> because FSL only runs a fixed effects analysis if it finds copes and
> varcopes, so zstats won't have corresponding varcopes (and they don't need
> them).  Secondly, if you use the first level zstat image, these are
> basically the Fisher's Z transformed correlations, which are distributed
> Normally with a sd of 1/sqrt(N-3), where N is the number of time points, so
> the variances are equal across subjects and runs.  The differences in
> variances have been adjusted for.  Instead, you should combine runs over
> subjects using OLS.
>
> Cheers,
> Jeanette
>
>
> On Wed, May 16, 2012 at 9:07 AM, Tom Johnstone <[log in to unmask]>
> wrote:
>>
>> Hi,
>>
>> Here would be the approach:
>>
>> 1. Run a 1st level analysis to compute connectivity maps for each
>> participant. The *ztstat* images will give a measure of connectivity
>>
>> 2. Separately for each subject, run a 2nd level fixed effects analysis
>> on the *zstat* images from the 1st level analysis. Specify a separate
>> EV for each session, and use contrasts of 1 -1  and  -1  1  to compute
>> connectivity differences between sessions (and 0.5 0.5 contrast for
>> the mean connectivity). The output *copes* will represent the
>> connectivity differences
>>
>> 3. Run a 3rd level analysis (either using Flame or Randomise) to
>> perform a 1-sample t-test on the *copes* from the 2nd level analyses.
>>
>> -Tom
>>
>>
>>
>> On Wed, May 16, 2012 at 2:36 PM, Maren Strenziok
>> <[log in to unmask]> wrote:
>> > Hi Tom,
>> >
>> > what would be the input to the fixed effects 2nd level analysis? Cope,
>> > zstats images, or feat directory?
>> >
>> > Maren