Brige,
The SureFit segmentation algorithm was originally designed for
segmenting T1-weighted MRI's of human cerebral cortex, plus other
volume data that have similar shape and image intensity attributes.
The fundamental requirements are that cortical gray matter be about 3
voxels thick and that white matter be brighter on average and CSF be
darker.
The hindbrain and cerebellum need customized handling that is
species-specific; this is also the case for the olfactory bulb in
rodents.
The segmentations of mouse and rat cerebral cortex that are the basis
of our rodent surface-based atlases
(http://sumsdb.wustl.edu:8081/sums/) were done by me personally several
years ago; it entailed a large amount of manual editing to achieve a
reasonable segmentation. I have not subsequently tried to segment
other rodent brains besides the two used for our atlas, nor am I aware
of anyone else who has done so using SureFit (or, for that matter, any
other algorithm).
I would be pleased to have the SureFit algorithm refined so that it
works adequately on rodent MRI's. This might well take some additional
effort on our part beyond just fixing the crashing problem that you
identified and Donna discussed in her email. Once this crash problem
is fixed (don't know how long that will take), we will let you know.
If you try SureFit/Caret again on the mouse, please keep us posted
whether it gives encouraging results or whether specific additional
problems arise.
David VE
On Jul 29, 2005, at 3:15 PM, Donna Hanlon wrote:
Hi Brige,
I am not someone who has had success (or failures) with mouse brain
segmentation, but I tried segmenting the map015 left hemisphere using
Caret, just to see what would happen. I made these selections:
* left the voxel size at 0.056mm
* GMpeak=45 ; WMpeak 60
* toggled off disconnect eye and skull and disconnect hindbrain, since
these have no chance of working on rodents
* toggled on generate segmentation
* toggled off fill ventricles, automatic error correction, and all
remaining options
The idea was to see if I could get an initial segmentation without
going any further. The progress bar showed it was working on inner
boundary (25%), but then Caret crashed with the following gdb stack
trace:
I knew there were crashing problems when toggling off disconnect
hindbrain, but I didn't know what would happen when the other options
were also toggled off.
We need to look at this and get it working. The idea is to replicate
the functionality SureFit had with these selections:
segment full volume
fill ventricles no
skip error correction
review surface first
skip sulci identification
But I don't know how long this will take. Meanwhile, if you're
running Linux, you might try using SureFit for this purpose, using the
settings above (after doing volume preparation -- orientation, setting
AC, cropping, and setting peaks).
Donna
On 07/29/2005 01:51 PM, Brige Chugh wrote:
Hi,
I have been trying to apply caret's automatic segmentation routine on
the mouse brain. I have registered and resampled the brain so that it
is similar to the map015 atlas and have performed n3 correction. I am
following the caret segmentation tutorial closely. I am able to
clearly discern the grey and white matter peaks but I have found that
when I try to run automatic segmentation, it always crashes. The
program closes and I get the message "segmentation fault". As I new
user to the caret environment, I would really appreciate any advice
from anyone who has had some success with mouse brain segmentation.
Thanks in advance,
Brige Paul
_______________________________________________
caret-users mailing list
caret-users@brainvis.wustl.edu
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
_______________________________________________
caret-users mailing list
caret-users@brainvis.wustl.edu
http://pulvinar.wustl.edu/mailman/listinfo/caret-users
