If such instability causes the estimation of the bias to become worse,
then the algorithm will revert back to the previous estimate for the
bias.
A possible cause of the problem is that there are too many Gaussians to
model the intensities of non-brain tissue. If non-brain tissue is
removed, then the non-brain class will contain very few voxels, and this
will increase the chance of a Gaussian collapsing into a delta function
with a variance approaching zero. If this happens, the algorithm could
become slightly unstable.
Best regards,
-John
-----Original Message-----
From: SPM (Statistical Parametric Mapping) [mailto:[log in to unmask]]
On Behalf Of Ged Ridgway
Sent: Tuesday, May 15, 2007 12:55 PM
To: [log in to unmask]
Subject: Re: [SPM] segment errror message
Hi Michelle,
John might comment further, but I notice that you are using SPM5's
unified segmentation (i.e. spm_preproc is called). This uses the
tissue probability maps under spm5/tpm/ for both segmentation and
normalisation, so your skull-stripped T1 image is not actually being
used at all.
I'm not sure why you'd get the warning message... check that you have
the latest updates for SPM5 (I don't think you do, based on the
warning reporting line 353...)
ftp://ftp.fil.ion.ucl.ac.uk/spm/spm5_updates
Check that you're skull-stripped images look okay and are reasonably
well-aligned with MNI-space (e.g. check reg with tpm/grey.nii). It is
just a warning and not an error though, so I guess if the results look
fine, it's probably okay.
Well, I hope that helps (and saves John at least a bit of time!)
Ged
> I'm running segmentation on skull-stripped brains, using a customized
> template that i plugged into the templates directory and renamed as
> T1.nii. the template is skull-stripped as well. only when i use a
> skull stripped template, i get this error message
>
> Running "Segment"
> Warning: Matrix is close to singular or badly scaled.
> Results may be inaccurate. RCOND = 1.289615e-20.
> > In spm_preproc at 353
> In spm_config_preproc>execute at 474
> In spm_jobman>run_struct1 at 1474
> In spm_jobman>run_struct1 at 1482
> In spm_jobman>run_struct1 at 1482
> In spm_jobman>run_struct at 1436
>
> - segmentation still runs through and seems fine, is this an error
> message i should be worried about in terms of the quality of the
> segmentation/vbm analyses?
>
> thanks,
> Michelle
|