I've encountered this before. In my case I believe it was caused by some abnormal edge voxels after co-registering a bunch of segmented scans, and the error seemed to propagate from the lower edge. It became very obvious when I double checked all the input scans (it was only one or two scans that caused it that were clearly different from the rest).

Try running this in matlab:

data=spm_select(inf,'any','Input RC images');%Select your dartel imports

for i=1:size(data,1);
tmp=deblank(data(i,:));
spm_check_registration(tmp);
pause(2);
end

As mentioned, the source of my error and the start of the propagation were in the same region. Looking at yours does make me wonder whether a single white matter segmentation has crept in with your grey matter inputs, or something similar (like a single sub-optimal grey matter segment with additional tissue due to WM disease for example).

Hopefully it will be something simple and easy to spot. 

- Chris


On Wed, May 9, 2012 at 5:27 PM, shahrzad kharabian <[log in to unmask]> wrote:
Thanks a lot.

I am using "Version 4290 (SPM8) 04-Apr-11" and I have exported "affine transformed" GM and WM from VBM8 toolbox "estimate and write" to DARTEL module (as it was suggested in VBM8 tutorial!). Could it cause the mentioned instability?!

Also, to Khader, the data is a mixture of healthy and non-healthy(in different stages of disease), old women. I have not checked the white matter lesions yet. So I have no idea about it and also subjects are rather in the same age range. BTW, thanks a lot, these are good points that I could look at them and search on them in future:)

cheers,
Shahrzad


From: John Ashburner <[log in to unmask]>
To: [log in to unmask]
Sent: Wednesday, May 9, 2012 5:36 PM
Subject: Re: [SPM] Dartel template

As Satra pointed out, the final template seems to be OK.  I therefore
think your overall alignment should be fine.

Which version of SPM8 did you use (and did you use more than the usual
GM and WM tissues)?  The problem is the result of an instability in
the way the template is smoothed, which I have occasionally seen.  As
far as I was aware though, I thought I'd fixed it in the more recent
SPM8 updates.

The smoothing is not the usual sort of smoothing that involves
convolving with a Gaussian.  Instead, the idea was to smooth by
including a regularisation model in the generation of the template,
which regularises more heavily in the earlier iterations and less in
the later ones.  Computations are only done with single precision
floating point, so it is a bit more difficult to ensure the various
computations are 100% stable (while still reasonably fast)

Best regards,
-John

On 9 May 2012 16:08, shahrzad kharabian <[log in to unmask]> wrote:
> Dear SPM experts,
>
> I have made a DARTEL template of about 68 brains of old subjects and I got
> the attached templates. It seems for me that the first 2 templates _0 and _1
> are rather strange. I need your help to know how should I deal with it, what
> the reason for this could be and if it might affect my further analysis (as
> the last template seems reasonably good to me).
>
> Bests,
> Shahrzad