Hi Alle Meije,
From the output, the affine normalization has gone badly wrong, which
has been an issue in the past (but much less so in the more recent
versions). I think in VBM12 Christian has a more extensive initial
matching approach to avoid this, but I would have to check where this is
implemented.
In any case, I think you have narrowed it down already to where things
go astray, namely the bias field correction. If the images are fomr a
single source, it seems more likely you have a bad apple than a bad
approach in general... Have you looked at a histogram of the offending
image, i.e. are there far outlying values that mess up histogram
matching? Is the datatype different from the other images? Does the
volume contain a lot of non-brain tissue, or large areas set to
0/NaN/Inf? If you lower the regulariazation further, does it look
better? As a failsafe, can you save a bias-corrected version and segment
that (just to see if it works then)? You could of course also try other
regularization schemes but MNI should of course be fine for the standard
priors and it would entail reprocessing all others...
Good luck,
Marko
Alle Meije Wink wrote:
> One of my 160 data sets (which were all approved for image quality)
> fails in the VBM segmentation of SPM12. I am using all the standard
> options, and as an extra I write the rc* images + warp fields.
>
> The scan looks normal although there is quite a clear bias field. Apart
> from the standard parameters
> I have tried different settings for the bias field regularisation
> matlabbatch{1}.spm.spatial.preproc.channel.biasreg
>
> Without regularisation the process ends with
> Warning: Matrix is singular to working precision.
> > In spm_get_closest_affine at 61
> ...
> Warning: Matrix is singular, close to singular or badly scaled.
> Results may be inaccurate. RCOND = NaN.
> > In spm_get_closest_affine at 67
> ...
> Failed 'Segment'
> Error using ==> svd
> Input to SVD must not contain NaN or Inf.
> In file "/home/data/amwink/matlab/spm12/spm_get_closest_affine.m"
> (v6137), function "spm_get_closest_affine" at line 68.
>
> With "matlabbatch{1}.spm.spatial.preproc.channel.biasreg = 0.0001" the
> process finishes -- I get the message "Done 'Segment' " -- but the
> output does not make sense. It looks as if the GM densities have been
> shifted and masked by a -differently shifted- standard GM mask (see PNG,
> c1(yellow) on the original).
>
> That made me think it could be a header issue but there is nothing
> special about that either.
> Attached are the scripts that I run for the Segment job and the header
> output, happy to send more.
>
> Does anyone know if this is fixable?
> Many thanks!
>
>
--
____________________________________________________
Prof. Dr. med. Marko Wilke
Facharzt für Kinder- und Jugendmedizin
Leiter, Experimentelle Pädiatrische Neurobildgebung
Universitäts-Kinderklinik
Abt. III (Neuropädiatrie)
Marko Wilke, MD, PhD
Pediatrician
Head, Experimental Pediatric Neuroimaging
University Children's Hospital
Dept. III (Pediatric Neurology)
Hoppe-Seyler-Str. 1
D - 72076 Tübingen, Germany
Tel. +49 7071 29-83416
Fax +49 7071 29-5473
[log in to unmask]
http://www.medizin.uni-tuebingen.de/kinder/epn/
____________________________________________________
|