Thanks a lot for your prompt answer.

>
>Hi
>
>On 26 May 2014, at 12:18, Mario Quarantelli <[log in to unmask]> wrote:
>
>> Hi,
>> I‘m following the DualRegression/UserGuide to run dual regression to analyse a set of resting-state studies (already normalized to the EPI template) to assess differences between the sensorymotor network  of patients (n=27) and normal subjects(n=18).
>> MELODIC results look fine and the sensorymotor network can be clearly identified (in this case IC 3).
>> However, the tmaps obtained during step 2 of dual_regression (I’m using the dual_regression script shipped with FSL 5.0) poorly overlap with the component I selected as SMN (their average shows indeed quite a large overlap with CSF).
>
>first thing would be to do a few sanity checks:   are you definitely looking at the right component number (eg - confusion can be caused by "timepoint" counting in FSLView starting at 0, whereas group-ICA web reporting starts counting at 1....?    Also, when you say that the group-average output from dualreg stage2 doesn't look right, this is in effect a group-average, not (eg) a two-group t-test difference output by the final randomise?

I checked and I had correctly selected the 3rd component (i.e. "fslroi melodic_IC melodic_IC3 2 1", and I checked with fslview the output).
The melodic_IC3 image file was then used for dual_regression (i.e. “dual_regression ../CMD_MELODIC.ica/melodic_IC3.nii.gz 1 SMN.mat SMN.con  5000  DR_3rd_IC `cat lista_files.txt`”)
I compared the actual averages obtained using the two approaches over the whole sample (patients+norms).
The average of the components provided by feat was obtained using:
fslmerge -t FEAT_SMN */stats/tstat*
fslmaths FEAT_SMN -Tmean mean_FEAT_SMN
to obtain the average of the components provided by dual_regression, I launched the following command in the directory (DR_3rd_IC) where dual_regression output was saved:
fslmaths DR_SMN.nii.gz -Tmean mean_DR_SMN

The first image looks like the SMN, the second one has higher values in the CSF.

>
>> On the other hand, if I use FEAT to run the second part of the dual regression (using as EV the timeseries from stage 1 of the dual-regression, i.e. dr_stage1_subject[#SUB].txt), the mean across all the subjects looks like the SMN, as expected.
>> In doing this I turned off all the options in feat (no FILM prewhitening, no spatial or temporal filtering and so on…).
>> Am I doing something wrong or actually FEAT is likely to perform better than fsl_glm for this task?
>
>Was the FEAT analysis using just the single relevant timeseres for the GLM model, or all components' timeseries?  If it's the latter, then this should look very similar to what you get our of fsl_glm (ie dualreg stage 2).

Both dual_regression and the FEAT analysis were run using only the relevant timeseries for the GLM model. 

>
>Cheers.
>
>
>> More in general, is there any specific reason why fsl_glm is used in the step 2 of dual_regression instead of using FILM?
>> Thanks for any help
>> mario
>
>
>---------------------------------------------------------------------------
>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
>---------------------------------------------------------------------------
>
>Stop the cultural destruction of Tibet
>
>
>
>