Hi Christine,
I'm guessing that you want to compare whole-brain resting state
networks between your groups. You should therefore be performing your
t-test on the dr_stage2_ic{I#C].nii.gz images.
But you don't need to do this manually - if you give dual_regression
your design matrix and contrasts, it will run the statistical analysis
for you. This is all covered in the FSL course practicals:
http://fsl.fmrib.ox.ac.uk/fslcourse/lectures/practicals/ica/index.html#group
So a more conventional pipeline would look something like this:
1. Preprocess the data for each subject and run them through group ICA
2. Set up a design matrix which encodes the statistical test you want
to run (you can use the Glm/Glm_gui tool for this)
3. Pass the group-ICA output, the original fMRI data for each subject,
and the design matrix+contrasts to the dual_regression script
Cheers,
Paul
On 05/05/2019, Christine Kindler <[log in to unmask]> wrote:
> Dear FSL experts,
>
> I'm newcomer in imaging and FSL. Actually, I bogged down with an FSL Melodic
> ICA analysis and do not know, how to carry on.
>
> I have one group (subject 1 to 12) with two sessions (before (“pre”) and
> after (“post”) treatment). I run MELODIC with my whole group data, including
> subjects before and after treatment in the concat-ICA mode. I got 134
> components, with 32 “true” networks after visual inspection.
>
> As next step I wanted to perform a paired t-test. As described on the fsl
> homepage (Single-Group Paired Difference (Paired T-Test)), I calculated the
> difference between the resulting components for subject 1 to 12 separately:
>
> fslmaths 001_post/filtered_func_data.nii.gz -sub
> 001_pre/filtered_func_data.nii.gz sub1_diff
>
> fslmaths 002_post/filtered_func_data.nii.gz -sub 002_pre/
> filtered_func_data.nii.gz sub2_diff
>
> etc…
>
> and subsequently used “fslmerge” to create a single 4D volume:
>
> fslmerge -t 4Dvolume sub1_diff sub2_diff sub3_diff sub4_diff sub5_diff
> sub6_diff sub7_diff sub8_diff sub9_diff sub10_diff sub11_diff sub12_diff
>
> Finally, I used the resulting “4Dvolume” to perform a one sample t-test:
>
> randomise -i 4Dvolume.nii.gz -o 4DvolumeT -1 -v 5 -T
>
> and used “fslmaths” to get the clusters and peak information from is the raw
> tstat image by masking the raw tstat image with the significant voxels from
> corrp:
>
> fslmaths 4DvolumeT_tfce_corrp_tstat1 -thr 0.95 -bin -mul 4DvolumeT_tstat1
> 4DvolumeT_thresh_tstat1
>
> As a final result I get one image with significant cluster across various
> networks, which I interpret as significant group differences “post versus
> pre across all networks”.
>
> Now I have two questions
>
> 1) I am still not sure, if this is statistical the right way to perform a
> Melodic ICA paired t-test and would like to hear your view on this – maybe
> there is a more appropriate way to perform the analysis?
>
> 2) If I perform the paired t-test by substracting the two sessions for each
> person separately, do I have to substract the raw
> “filtered_func_data.nii.gz” images or do I have to perform first a dual
> regression analysis and then substract the “dr_stage2_subject[#SUB].nii.gz”
> images?
>
>
> Thanks in advance.
>
> Kind regards,
>
> Christine
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
>
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|