This is probably related to (if not the same as) T11477 . For reference this seems to be a tilted gantry image, is this correct?

I tested this issue on the current master and on the Diffusion 2013.09 release. On the current master the problem is as reported, on the 2013.09 release pressing reinit is still broken, but you will not need to do a reinit before starting the segmentation.

Reinit should not let the image disappear. In this case the reinit even breaks the geometry of the renderwindows. You can see this in the 3D windows, all planes seem to be at the image border after pressing reinit.
Was this image somehow modified?
We had once such a case but we had the assumption that this image seemed to be mirrored.

3D Image to world matrix of the attached image is roughly:

1.8 0.1 0.8
0 -1.7 1.6
-0.3 0.5 5.2

The image was directly exported from the scanner (Siemens Trio) and converted to NIfTI using dcm2nii (http://www.mccauslandcenter.sc.edu/mricro/mricron/dcm2nii.html). No further processing steps were done.

I have another (T1) image where this happens.

Matrix is:

0 0 -1
1 0 0
0 1 0

As this does not only happen with diffusion images I am moving this to core and increasing severity/importance.

Do you have any updates? Just compiled the current Git master, but unfortunately this problem still exists. Will it be fixed in the next release? Do you need more test cases?

There was once a bug regarding a similar issue. I think the discussion in this bug is relevant for this one, see T11477

octave/methlab/matlab:>
itwtm=[1.8, 0.1, 0.8; 0, -1.7, 1.6; -0.3, 0.5, 5.2];
det(itwtm)
ans = -17.808

lefthanded coordinate orientations are not well liked by 'Reinit',
I think because of ensurePerpendicularNormal().

I just pushed bug-11477-lefthanded-and-righthanded-geometries,
which seems to correct this bug.

Please test the hell out of it and report results.