Print

Print

Chirag, another clarification:


the IC volume from melodic should be binarized before it is used as a mask in randomise. 


ex:

cd ${melodic_dir} # the directory where you performed melodic group ICA
fslsplit melodic_IC.nii.gz 

extension='_bin.nii.gz'

fslmaths ${melodic_dir}/vol000${component} -bin ${melodic_dir}/vol000${component}${extension}


for component in 2 3 5 7 8 
do
randomise -i dual_regression/dr_stage2_ic000${component}.nii.gz -o randomise/comp$component -d file.mat -t file.con -f file.fts -m ${melodic_dir}/vol000${component}${extension} -x -T --uncorrp
done


Best,

Dina


--

Doctoral Candidate

Brain Connectivity and Cognition Laboratory

Cognitive Neuroscience

Department of Psychology

University of Miami

From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Dajani, Dina R <[log in to unmask]>
Sent: Thursday, October 12, 2017 6:07:16 PM
To: [log in to unmask]
Subject: Re: [FSL] Nuisance removal post group ICA
 

There was a hidden step I forgot to include... see below (fslsplit). The mask used in randomise is the IC of interest (created during the group ICA).


Also I like to output many different options for randomise (the -x and -uncorrp flags) just in case later I choose to look at them. If you don't output them now, and later you decide you want these files, you will have to redo randomise all over again. The default num of permutations for randomise is 5000, so the -n 5000 flag is unnecessary.


##melodic group ICA

melodic -i ICAinput.txt -o ICAresults_15 -a concat -d 15 --report --tr=2.5 -v
#dual regression runs on all ICs
dual_regression ${melodic_dir}/melodic_IC.nii 1 ${new_dir}/${analysis}.mat ${new_dir}/${analysis}.con 0 ${new_dir}/dual_regression `cat ${new_dir}/ICAinput.txt`

##hidden step I forgot to include
cd ${melodic_dir} # the directory where you performed melodic group ICA
fslsplit melodic_IC.nii.gz #this splits the 4D melodic file into volumes. each volume is a component. choose the ones of interest for randomise (e.g., 2 3 5 7 8 in this example)

#run randomise ONLY on the ICs of interest (loop through the components)
for component in 2 3 5 7 8 
do

randomise -i dual_regression/dr_stage2_ic000${component}.nii.gz -o randomise/comp$component -d file.mat -t file.con -f file.fts -m ${melodic_dir}/vol000${component} -x -T --uncorrp
done

Best,
Dina



--

Doctoral Candidate

Brain Connectivity and Cognition Laboratory

Cognitive Neuroscience

Department of Psychology

University of Miami

From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Chirag Limbachia <[log in to unmask]>
Sent: Thursday, October 12, 2017 5:11:02 PM
To: [log in to unmask]
Subject: Re: [FSL] Nuisance removal post group ICA
 
Thanks again Dina,

I have couple of things to clarify. 

I went through the dual_regression script and I found they use the following settings for randomise:

randomise -i $OUPUT/dr_stage2_ic$jj -o $OUTPUT/dr_stage3_ic$jj -m $OUTPUT/mask $DESIGN -n 5000 -T -V

And you suggested the following:

randomise -i dual_regression/dr_stage2_ic000${component}.nii.gz -o randomise/comp$component -d file.mat -t file.con -f file.fts -m ${melodic_dir}/vol000${component} -x -T --uncorrp

Q1) Is the vol000${component} the mask that is created in the dual_regression directory when running stage 1 and stage 2 using dual_regression? and does it remain the same for all ICs? 

From the code you provided, it looks like the mask (vol000$...) updates as the loop iterates. I checked my melodic directory, but did not find the vol000$... images. Should I just use the mask image that is present in the dual regression directory for all ics of interest?  

Q2) Is the setting '-x -T --uncorrp' more appropriate for this kind of ic testing than '-n 5000 -T -V' mentioned in the original dual_regression script?

Your help is very much appreciated.

Thank you!
Chirag


On Thu, Oct 12, 2017 at 12:15 PM, Dajani, Dina R <[log in to unmask]> wrote:

Here is some sample code:


##melodic group ICA

melodic -i ICAinput.txt -o ICAresults_15 -a concat -d 15 --report --tr=2.5 -v

#dual regression runs on all ICs
dual_regression ${melodic_dir}/melodic_IC.nii 1 ${new_dir}/${analysis}.mat ${new_dir}/${analysis}.con 0 ${new_dir}/dual_regression `cat ${new_dir}/ICAinput.txt`

#run randomise ONLY on the ICs of interest (loop through the components)
for component in 2 3 5 7 8 
do
randomise -i dual_regression/dr_stage2_ic000${component}.nii.gz -o randomise/comp$component -d file.mat -t file.con -f file.fts -m ${melodic_dir}/vol000${component} -x -T --uncorrp
done



--

Doctoral Candidate

Brain Connectivity and Cognition Laboratory

Cognitive Neuroscience

Department of Psychology

University of Miami

From: FSL - FMRIB's Software Library <[log in to unmask]> on behalf of Chirag Limbachia <[log in to unmask]>
Sent: Thursday, October 12, 2017 5:28:43 AM
To: [log in to unmask]
Subject: [FSL] Nuisance removal post group ICA
 
FSL Experts,

I ran a group ICA with free estimation and got 78 ics. Of which, only 29 look like plausible RSNs. I am told that I should run stage 1 and 2 of dual regression and then only use the 29 ics for group comparison in the third stage using randomise or PALM.

My questions is,

How do I select only the 29 out of 78 ics as input maps for randomise/PALM?

Thank you,
Chirag


Sent from my iPhone