This is nothing to do with SPM's segmentation.  When you set up the statistics, you specified which voxels to analyse via the "Masking threshold" (which probably used its default value of 0.8).  The NaNs are the voxels in the beta images that are not defined because the beta values were not computed.

Best regards,
-John


On 9 June 2016 at 11:35, Louis Thibault <[log in to unmask]> wrote:
Dear SPMers,

I'm faced with a bit of a strange problem.  I'm attempting to select voxels in my beta-maps with on a mask that I've created on the basis of a contrast.  However, upon selecting said voxels, I discovered that they all contain NaN values.

Intrigued, I loaded up the beta maps using spm_check_registration and found, much to my chagrin, that SPM's segmentation function has apparently removed brain matter around the cerebellar tentorium (please see attached).

What could be responsible for this problem?

To reiterate, the masks that led to the discovery of this problem are based on first-level contrasts, which worked beautifully.  Strange, considering there's a gaping hole in the beta-map  ... no?

Many thanks in advance,
Louis
Inline image 1
--
Louis Thibault
Laboratoire de Psychologie de laPerception
Room H418B
45, rue des Saints-Pères
75006 Paris