I usually generate a brain mask from my high-resolution anatomical brain in
standard space, dilate it, resample it to my functional resolution, and then
apply it to my function data in standard space before running melodic. I do
this separately for each subject. I run melodic with --nobet and --nomask. I'm
actually curious myself as to whether or not this is a good way to do things.
No matter what you do you're likely to end up with a few non-zero voxels just
outside the brain. I think it's better to include a few non-brain voxels than
it is to mask out voxels from the cortical surface.
On Thursday, October 20, 2011 08:38:45 you wrote:
> Hi there,
>
> I am running the melodic command without masking since the images are
> already preprocessed (betted) and are in standard space. See command
> example below:
>
> melodic -i melodic_input_subject_ADOs_new_line_NMI.txt -o
> melodic_ados_all_NMI -a concat --nomask --nobet --mmthresh=0.5 --tr=3
> --report -v --Oall
>
> I noticed that the brain extracted fmri data (the warped and the ones in
> native space) are non-zero outside the brain resulting in analyses being
> carried out outside the brain.
>
> I looked at the masks that I got from feat preprocessing, and they are
> bigger that the brain, matching the non-zero space outside the brains. So
> I can't really use those, and the MNI mask may not match our population.
> Manual thresholding is also not a good idea I guess.
>
> So first, is this a statistical disadvantage (ie more voxels to compare)?
> and has anyone have had similar problems and how to deal with them.
>
> Thanks very much in advance!
>
> Torsten
|