Dear FSL experts,
the discussion below has been very useful for me but I still have some doubts:
1) After running dual_regression without randomise on the resulting melodic_IC.nii.gz maps, I get 3 dr_stage2 outputs, I guess the correct ones to subtract are the "dr_stage2_subject[#SUB].nii.gz" as described here http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/DualRegression/UserGuide. Is that right?
If so, how would you suggest to proceed when the repeated measures are 3 (longitudinal design) on the same group?
2) After using fslmaths and fslmerge in this post it is said that a 3-group one-way ANOVA (in this case between subject design) is possible.
I suppose this would be done using randomise at this point and setting up the F contrasts, how should I specifically set the GLM in this case?
3) Is it better doing it with an ANOVA or with 3 separate t-test?
I hope my questions are not too simple and that everything is clear.
Just to specify: I have both repeated measures and between groups design analyses to run but not together (one analysis is within and one separate is between groups, both wiith 3 groups of measures).
Thank you very much in advance for your help,
Francesca
#################################################
Hi Alain,
Please see below:
On 21 May 2015 at 11:01, Alain Imaging <[log in to unmask]> wrote:
Hi everybody!
I am totally green to MELODIC and I would like to have some hints.
I have performed a longitudinal study with three groups. Each group received a resting state fMRI before and after a treatment. Our hypothesis is that the three different treatment affect the resting state functional connectivity in a different way. So what I thought to do is performing MELODIC (multi-sessions temporal concatenation) on my functional data, in order to perform dual regression as a second step and comparing the three groups.
Here come the questions:
1) Since I have two session for each subject, that are identical as for what the subject does (i.e. nothing) but should differ because of treatment, what should I do ? Should I nonetheless temporally concatenate the two sessions and feed one 4D image for each subject ? Should I feed the two sessions as separate images (i.e. enter subjects x 2 images) ?
You can use both files of all subjects into a single Melodic run, with temporal concatenation. There should be no biases on the IC maps.
2) I thought that I should run melodic on the whole sample, without dividing it by group. Am I right ?
Yes. Try to use groups of the same size, using the same logic as with study-specific templates as in, e.g., VBM.
3) Related to question 1. If I feed MELODIC with the temporally concatenated session for each subject, how I can get back and observe the effect of time ? And if I feed MELODIC with two sessions for each subjects separately, can I calculate then a delta between the component at time 1 and time 2 and use this delta for testing differences in the group using randomise ?
If you use the dual regression, you'll end up having one spatial map per component, per session, per subject. Then it's possible to subtract session 1 from session 2 for each component, and run a 1-sample t-test in randomise. This will tell something about time changes for each of the components. It's also possible to do a 3-group one-way ANOVA on the subtracted images.
Hope this helps.
All the best,
Anderson
I hope that my questions are clear enough.
Thank you in advance
Alain
|