Hi Vincent,
Not exactly what you want, but BrainWeb has models for the skull (and
other things besides WM,GM,CSF) that it uses for simulating new
images. I believe these came from semi-supervised segmentation (e.g. a
Neural Net classify with manually picked points) followed by some
editing, but they are not perfect.
Take a look at:
http://www.bic.mni.mcgill.ca/cgi/brainweb1?alias=subject43_skl
and the other subjects and tissue-classes. The "skulls" don't look
perfect to me (isn't that a nose?!), and the other classes have things
which I think might be dodgy (GM around the brainstem, etc.), but my
neuroanatomy knowledge is pretty basic.
There are also now 21 BrainWeb images in total, so you could attempt
to normalise them (e.g. use SPM5's segment on the T1 data, then
normalise-write the models) and average to create rough priors. Note
though that 21 is probably not enough for meaningful priors, as
biological variability is very significant (imagine trying to model
the appearance of a random person's face just by aligning it to an
average of a few dozen different faces). I think the ICBM tpm images
used something like 452 subjects...
I wonder if Alex Hammers attempted to estimate the skull for his
hand-segmented data? Again, probably not enough subjects for priors,
but probably better segmentations than BrainWeb...
Hope that helps (at least a tiny bit),
Ged.
Vincent Keereman wrote:
> Hello,
>
>
>
> I’m segmenting MR brain scans, but apart from WM, GM and CSF, I also
> need to locate the bone structures correctly. Does anyone know if there
> are any tissue probability maps available for bone?
>
>
>
> BR,
>
>
>
> Vincent
>
|