> The question is how accurately should one make the segmented image in order
> for SPM5 works well. Currently, I have a vervet monkey template (an
> averaged skull-scripted brain T1 image).? I am thinking about segmenting it
> with FSL to make our own templates for the segmented images.
This depends on many factors. If you have just a single image segmented into
different tissue classes, then you should smooth the tissue class images
spatially so that the smoothed version more closely reflects probabilities
for a population of subjects. I'm not sure what the best amount to smooth it
by is though (another empirical issue).
Best regards,
-John
> -----Original Message-----
> From: John Ashburner <[log in to unmask]>
> To: [log in to unmask]
> Sent: Tue, 19 Feb 2008 10:35 am
> Subject: Re: [SPM] reg. Mouse brain extraction and segmentation
>
> > I am trying to perform Mouse brain extraction and then
> > segmentation on T2 weighted 256 x 256 x 15 16 bit images but not able to
> > find the corresponding resources. Can anyone please guide me to achieve
> > my aim?
>
> In order for a segmentation algorithm to work, it needs some idea of what
> the tissues are that you hope to identify. Sometimes, the approach
> involves "supervised learning", whereby a few voxels are labled as being of
> various tissue classes. The algorithm then tries to determine a strategy
> to predict these lables from information in the images.
>
> The strategy in SPM is kind of semi-supervised, whereby maps that reflect
> the probability of different classes are used. The tools in SPM are
> unlikely to work unless you can track down tissue probability maps for
> mouse brains. There may be other segmentation algorithms out there that do
> not need tissue probability maps. Here are a few places to look:
>
> http://idoimaging.com/index.shtml
> http://www.cma.mgh.harvard.edu/iatr/
> http://www.nitrc.org/
>
> Best regards,
> -John
>
>
>
>
>
>
> ________________________________________________________________________
> More new features than ever. Check out the new AIM(R) Mail ! -
> http://webmail.aim.com
|