Dear Benjamin,
As the resolution and contrast are worse in functional images
by comparison with structural images then two things happen:
(a) the brain extraction is less accurate; and (b) the registration
is more influenced by any errors in the brain extraction. As a
consequence mcflirt, as run on functional images, works much
better without brain extraction. By comparison, normal T1-weighted
high-resolution images benefit from brain extraction as it often
removes large portions of neck (also rarely imaged in functionals)
that improve the robustness of anatomical registration (with flirt)
without sacrificing accuracy, since the features within the brain
are strong enough to make the registration insensitive to small
errors in the brain extraction.
So please don't run brain extraction on your functionals prior to
using mcflirt.
All the best,
Mark
On 23 Aug 2011, at 20:55, Benjamin Kay wrote:
> I'd been considering doing something like this:
>
> mkdir func func-brain
> fsl4.1-fslsplit func.nii.gz run/func- -t
> for FILE in func/*; do fsl4.1-bet "${FILE}" "run-brain/${FILE##*/}"; done
> fsl4.1-fslmerge -t func-brain.nii.gz func-brain/*
> fsl4.1-mcflirt -in func-brain.nii.gz -out func-brain-mcf.nii.gz
>
> However it's been mentioned twice before on this mailing list (by Steve Smith) that one should not perform brain extraction before invoking mcflirt.
>
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;1f232570.05
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=FSL;9b7f3193.0906
>
> On the other hand, the FAQ recommends doing brain extraction on both volumes before running flirt.
>
> http://www.fmrib.ox.ac.uk/fslfaq/#title_flirt
>
> Given that flirt and mcflirt are ostensibly both doing linear registration, why is brain extraction helpful to one but harmful to the other?
>
|