It would need a bit of empirical exploration to say what would give the most
accurate segmentations. I have found one drawback with the new segmentation,
which relates to the lack of tissue probability maps for eyeballs. This
causes the algorithm to try to squash eyeballs into the CSF class, because
aqeous/vitrous humour has a signal intensity very similar to that of CSF. If
you have two channels of data, then that provides more evidence that eyeballs
should be treated as CSF, so the algorithm tries even harder to squash them
into the brain. Therefore, if you go multi-spectral, you may wish to
increase the warping regularisation slightly. This would prevent the
nonlinear warping from deforming the data too much.
If you want to achieve more precise spatial normalisation, then you would
probably use DARTEL. If you go via the route of using the new segmentation,
then make sure that the segmentation is set to generate "imported" GM and WM
images (because the DARTEL import can not deal with the new seg8.mat files).
I would also suggest you generate native space GM, which can be warped later
using the option to spatially normalise images to MNI space.
There still appears to be a slight glitch in the option for normalising to MNI
space, but this only shows itself when "imported" data are not generated
using the default settings for voxel size or bounding box. I will fix it
soon(ish) though.
Best regards,
-John
On Wednesday 15 April 2009 11:42, Cyril Pernet wrote:
> Dear John,
>
> I'm starting a new project and before acquiring data I was
> wondering what would be the best things to acquire for segmentation?
> Aiming to do a VBM analysis I'll acquire the usual hight res T1 -
> would it make sense to also get hight res T2 or something else to get
> a better segmentation/normalization with SPM8?
>
> Cheers
>
> Cyril
|