Hi Alex,
If you have a lot of neck in your images then BET doesn't work
well unless you either truncate the images to remove most of
the neck (using avwrio for instance), or select the coordinate
of the centre of the brain (using FSLView for instance) and
specify this with the -c option in BET. Note that little bits of
non-brain tissue are fine but substantial amounts do affect
the registration.
As for registration error - it depends on what images you are
registering. For EPI to structural in the same individual, you
expect to get within 0-2mm for most areas in a reasonable
registration (remember that your EPI voxels are often 4mm)
but this gets worse if you have substantial B0 (susceptibility)
distortion. For structural to standard space, that is much harder
and because of population variability the result depends a
lot on the area and how "normal" the individual was. Again
0-2mm would be a good registration, but around the ventricles
it may be more than this. Of course all of this is specific to
affine registration to standard space (the only option in FSL
currently).
Hope this answers your questions.
All the best,
Mark
On Wednesday, June 2, 2004, at 08:34 am, Alex Fornito wrote:
> Hi all,
> I'm a first time fMR researcher and FSL user.
> Just had a couple of questions re: registration as implemented in FEAT.
>
> 1. I'm following the recommedations to register the epi to the template
> via the hi-res structural. How well stripped (using BET) does the
> structural have to be? I am often left with residual bits of non-brain
> tissue (especially neck) and I was wandering whether this would greatly
> affect the registration results for my purposes (ie., epi to template).
> Would manually removing non-brain tissue be worthwhile? (playing with
> FIT
> and g still doesn't yield great results in many cases, unless I'm
> doing it
> wrong?)
>
> 2. Are there any rules of thumb re: determining what's a minimally
> acceptable registration? Are minor deviations around the edges ok,
> provided there isn't evidence of gross misalignment? How large would
> these
> minor deviations have to be before they are considered problematic? Are
> there any examples of bad, or (just) decent registrations on your
> website?
>
> Many thanks for your help - am enjoying the software!
> Alex
|