Hi
> On 14 Jun 2015, at 04:18, Oihane Rilo <[log in to unmask]> wrote:
>
> Dear FSL experts,
>
> I would like to perform a tensor independent components analysis (ICA). However, I have several doubts about it and I would be very grateful for some feedback.
>
> 1) I have noticed that Melodic can be run in the FEAT gui as well as in the Melodic gui. However, the available options related to Melodic in each gui are different. So, is there any case in which the use of one gui could be a better option than the other?
>
Both run the same command line tool, the option simply exists in the FEAT gui to make it easier for people to run first-level ICA. For group ICA you should use the Melodic GUI or the command line after preprocessing
> 2) In the MELODIC gui, in the registration section, it is included the option “Resampling resolution”, which by default is 4mm. The voxel size of my fMRI and T1 images is 1.67 x 1.67 x 3.8 and 0.977 x 0.977 x 0.6, respectively. Is the default resampling resolution adequate for my images?
Well, it’s simply there to allow you to adjust your analysis in cases where you don;t have lots of memory available on the comp nodes. The easiest setting is to use ‘2mm’ in that it will permit loading the output into fslview and compare IC components to the different atlases available. Note that we would typically recommend isotropic voxel size acquisition
>
> 3) In the MELODIC gui, in the data section, the default value for the high pass filter cutoff is 100. I am working with an event-related task, it would be better to adjust that value depending on my block design total cycle time?
>
You can check with the FEAT guy what the different hp filter settings do to your specific design.
> 4) In the MELODIC gui in the Post-Stats section, it could be included the timeseries and session/subjects design matrix and contrasts. My study includes two groups (experimental/control) and two MRI acquisitions (pre-treatment/post-treatment). I am interested in test the interaction effect, so I suppose that a 2 way mixed effect Anova would be the best option. Please correct me if I am wrong.
Sounds like standard paired t-test
>
> a) Where I have to include the .mat and .con files in the timeseries or session/subjects option?
>
in the session/subject design
> b) I would like to verify if the following set-up appropriate for my analysis (Time: pre-treatment= 1; post-treatment= -1// Group: experimental= 1; control= -1)?
>
>
use the GLM gui to create the default paired-t test design and contrast matrices.
> Group Time GroupxTime Subj01 Subj02 Subj03 Subj04
> Input1 (subject01 exp-pre) 1 1 1 1 0 0 0
> Input2 (subject01 exp-post) 1 -1 -1 1 0 0 0
> Input3 (subject02 control-pre ) 1 1 -1 0 1 0 0
> Input4 (subject02 control-post) 1 -1 1 0 1 0 0
> Input5 (subject03 exp-pre) 1 1 1 0 0 1 0
> Input 6 (subject03 exp-post) 1 -1 -1 0 0 1 0
> Input7 (subject04 control-pre) 1 1 -1 0 0 0 1
> Input8 (subject04 control-post) 1 -1 1 0 0 0 1
>
>
> C1 1 0 0 0 0 0
> C2 0 1 0 0 0 0
>
>
> Finally, I would like to run 2 way mixed effect Anova in DTI. Is there any way to use this matrix or a similar one?
>
You can use TBSS
hth
Christian
> Thank you in advance,
>
> Oiane
|