Dear Clare,
The segmentation in FIRST for the cerebellum
is based on manually-labelled data and this is
what it is trying to replicate. I've included an image
below to give you an idea of the detail - which does
not include getting reliably inside the sulci. If you really
want to try and separate CSF from non-CSF (which is
going to be quite tough given the amount of partial
volume which exists in the cerebellum) then I would
take the FIRST segmentation and mask it with a
FAST segmentation obtained from the whole brain.
If you exclude CSF found by FAST then you will get
some improvement, but it is still far from actually
delineating the individual sulci of the cerebellum.
I definitely would not recommend the threshold
option as it will be difficult to define a sensible
value given that it will only use intensities inside
the cerebellum to define the mean and standard
deviation, which will include a lot of partial volume
voxels and hence not be very useful. Running FAST
on the whole brain will allow it to estimate the
WM, GM and CSF mean intensities in a much more
stable and reliable way.
All the best,
Mark
On 3 Aug 2010, at 10:11, Clare Gibbard wrote:
> Hi Mark,
>
> Thank you for your reply. The reason I am interested in using the
> thresh boundary correction option is because none of the cerebellum
> segmentations I have run using the fast boundary correction option
> are completely defining the cerebellum - they are including the CSF
> of the sulci. Therefore I am trying to work through sensible
> options for refining the segmentations and I thought that one of
> those could be altering the boundary correction option.
>
> I know that fast is the preferred method of boundary correction for
> the cerebellum, but I thought I would try the thresh option and see
> how it affects the segmentations. However, if it is not recommended
> I might try to think of other options for improving the outline.
> Thanks for clarifying it for me.
>
> Kind regards,
> Clare
>
>
|