Hi Olivier,
I've CCed the list, as I'm not too sure about this, so a second
opinion might help...
> The Fuzzy datasets have many tissues. What combination do you use to
> compare to the three classic WM, GM and CSF probability maps obtained
> after segmentation?
I think some papers that I've seen have just done the obvious thing of
comparing each SPM segmentation with the equivalent fuzzy model. I
guess your concern is that in SPM GM+WM+CSF covers the entire brain,
so one or more of the three "tissues" might be more than one
real/BrainWeb tissue.
I've had a quick look at the other tissues in the online viewer. I
think all of them except glial matter are outside the brain, e.g. they
are things like dura or scalp, which should not be included for
comparison to SPM's segmentations.
I gather that glial cells are non-neuronal cells present throughout
the brain, which makes it sound like a meaningful glial segmentation
might not exist anyway. In practice, in BrainWeb, the glial matter is
between the ventricular CSF and the surrounding white-matter:
http://www.bic.mni.mcgill.ca/cgi/brainweb1?alias=phantom_1.0mm_normal_gli
Since they are non-neuronal, one might argue that you should lump them
in with the WM class, if you wanted to include them. I guess the only
disadvantage with not including them would be that an apparently
"perfect" SPM segmentation of the GM and WM would not be possible, but
this is unrealistic anyway. So I think I'd simply compare grey with
grey, white with white, etc, and ignore glial matter.
Because the partial volume effect gives tissue between CSF and WM an
intensity quite similar to GM in T1 MRI, programs like SPM tend to
classify the CSF/WM boundary as GM (there has been some discussion
about this on the list before). So I think SPM would almost certainly
classify BrainWeb's glial matter as grey.
Note that the new BrainWeb 20 models don't include the glial class any
more. They do however add another tissue within the brain, vessels:
http://www.bic.mni.mcgill.ca/cgi/brainweb1?alias=subject04_ves
Again, I think it would be simplest (and probably best...) to ignore
this class, rather than to try to combine it with grey or white.
> Also, there are two WM volumes: white_1mm and white_ partial_1mm. I
> think the second one is the good one to use.
I'm not sure what you mean there... I can see only the one (fuzzy) WM
class, which looks fine to use:
http://www.bic.mni.mcgill.ca/cgi/brainweb1?alias=phantom_1.0mm_normal_wht
If you have two WM volumes I would compare them both to this one.
Best,
Ged
|