Hi - just a couple of add on points.
It is probably the case that the bias field is the problem here. I
know FAST can deal with a bias field, but
I suspect it is tuned for smaller gradients than the one you have.
Was the data acquired with a surface coil?
Anyway, if this is true, there are several things you might try, but
they may need a bit of playing around.
1. Significantly increase the number of iterations in fast (-i)
( the default settings are tuned for the kind of brains Yongyue was
used to seeing..)
2. Significantly reduce the bias field smoothness in fast (-l) - this
will allow fast to estimate quickly changing bias fields - I suspect
you should always do this in macaques because the brain is smaller
and therefore the bias field will be less smooth (when measured
absolutely).
3. Now we are really getting desperate - divide each slice by its
mean value - you might need to do something a bit different for the
end slices that are mostly grey matter, but this will certainly have
a dramatic effect.
Hope these things are slightly useful
T
On 1 Feb 2008, at 02:57, Matt Glasser wrote:
> Joe,
>
> There are probably several things going on here: 1) Motor cortex
> contains
> more myelin than other cortical areas, and as a result, T1 contrast
> is not
> as good as other cortical areas. 2) You should use the -or and the
> -ob
> switches of the fast utility to check that the bias correction is not
> actually making things worse at the top of the brain. I have found
> in some
> instances that it helps to set a lower threshold to remove any low
> intensity
> regions around that outside of the brain that were not removed by
> bet. 3)
> If neither of these methods work, you can improve the tissue
> contrast by
> using fslmaths to square the image, which increases the difference
> between
> low and high values. You may have to set an upper threshold to
> remove blood
> vessels, which will increase in intensity much faster than other
> regions, to
> the point that they segment in a separate category.
>
> Peace,
>
> Matt.
>
> -----Original Message-----
> From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> Behalf
> Of Joseph T. Devlin
> Sent: Thursday, January 31, 2008 6:20 PM
> To: [log in to unmask]
> Subject: [FSL] Tissue segmentation in macaques
>
> Hi FSLers,
>
> I've been trying to use FAST to get a good grey/white segmentation of
> some monkey brains and having been running into some difficulties.
> FAST does a nice job with the ventral half of the brain but more
> dorsally, it is classifying most voxels as white matter. Although by
> eye, there is a clear grey-white demarcation, in the more dorsal
> areas the grey matter values appear to be closer to WM values,
> especially WM values in the ventral part of the brain. I thought
> this might be a bias field issue, but the bias field outputted by
> FAST definitely does not show a ventral-dorsal gradient. To be
> honest, I'm not sure how this is generated, so I'm not certain what
> to make of that.
>
> Regardless, the information necessary to successfully segment GM from
> WM appears to be present in the image -- at least to my eye, the
> differences are clear and robust throughout the entire
> image. Nonetheless, I can't seem to find a way to robustly extract
> this with FAST and would welcome any suggestions.
>
> Thanks, in advance.
>
> Joe
>
|