The registration part of the new segmentation needs to compute the
gradients of the logs of the tissue probability maps. To ensure a nice
stable registration, these should be spatially smooth and the values
should not be -Inf (log(0)=-Inf).
I used the spm_dartel_smooth.m function to smooth tissue probability
maps (actually, the sufficient statistics for computing them are passed
to the algorithm). This is described in my NeuroImage paper on
computing average shaped tissue probability maps.
To be on the safe side, the bit of segmentation code that deals with
loading these tissue probability maps adds a tiny value before computing
the logs (line 55 of spm_load_priors8.m) and also includes a bit of
extra spatial smoothing (line 58 of spm_load_priors8.m ensures that the
b-spline interpolation is slightly higher degree than it needs to be).
Best regards,
-John
Hi John
Just a quick question regarding the tissue probability maps ---
you
note that the probabilities should never be exactly zero or
one. We
had some issues of missegmentation of soft tissue and bone from
the
neck being classified as gray matter (for T1 scans); setting the
GM
probabilities for these clear non-brain regions to 0 seemed to
significantly improve the situation. Should we not have done
that?
Best regards,
Jonathan
On Thu, Jul 8, 2010 at 3:10 PM, John Ashburner
<[log in to unmask]> wrote:
> I don't envisage any specific problems relating to Dartel, but
the
> results are likely to depend on how well the CT data can be
segmented.
> This will depend on the tissue probability maps used for the
new
> segmentation, so you may need a bit of trial and error to
generate
> these. Note that the probabilities in these should never be
exactly
> zero or one.
>
> Best regards,
> -John
- Show quoted text -
>
>
>
> I'm interested in using DARTEL to determine deformation
> parameters on non-brain data, specially cardiac CT, as
it is fit
> to a template. Other than needing a cardiac CT
template, does
> anyone foresee any problems in using DARTEL for cardiac
CT?
> -Greg
>
> --
> John Ashburner <[log in to unmask]>
>
--
John Ashburner <[log in to unmask]>
|