EPI data are pretty distorted and have a lot of dropout.  Essentially, if the algorithm gets an accurate rigid body registration of one part of the image, then it's likely to misregister another part.  This is simply because the two images are different shapes and therefore can not be accurately aligned via a rigid-body transform.  Take a look at the inferior parts of the frontal lobes for example.

Best regards,
-John



On 10 October 2013 15:09, H. Nebl <[log in to unmask]> wrote:
Dear SPM experts,


sometimes in the past I had the impression that the automatic coreg step tended to be supotimal along z axis for some subjects. When directly comparing EPI and anatomy, for example the ventricles were already visible in one of them, but not in the other.

Interestingly this also happens when EPI and anatomy seem to be aligned quite well before they enter the automatic coreg step. Some reasons that came to my mind:
1) For EPI the slice thickness is generally poorer than the in-plane resolution, so if there is any displacement, it should be more obvious along z.
2) The EPI often does not cover parts of the brain which are present in the anatomy, structures like brain stem/cerebellum. So the displacement might be related to different "centers of mass" (?). Under normal circumstances this should also occur more frequently along z. Using a skull-stripped version of the anatomy typically helped in these cases and minimized the displacement.


Now I came across an old data set recently. The EPI resolution was 3.75 mm in plane with a slice thickness of 2.5 mm. The resolution of the anatomy was 1.09 x 1.1 x 1.09 mm. At first I manually reoriented the structure so that it roughly matched the orientation of the first functional volume (just translations, no rotations). Based on the structural volume I reoriented all the data to AC-PC in the second step and then went on with standard preprocessing.

When checking the automatic coreg I noticed that the displacement of the structural volume along z seems to be higher after the coreg (see ventricles in the attached picture, SPM's interpolation viewing option disabled). I then tried with a skull-stripped version (structure masked with (GM+WM)>.5 or (GM+WM+CSF)>.99), but more or less the same results.

Basically the displacement tends to be about 2-4 mm, so corresponding to the slice thickness of the EPI, but not all of the subjects are affected. Of course the displacement is rather small, but nonetheless I would like to minimize it if possible (say, if you're interested in basal ganglia of if you want to mask CSF voxels).Maybe I'm just hyercritical, dunno :-)

Did you encounter similar things? And what would you suggest?  I could reorient the structures after automatic coreg, but this is a very subjective solution in the end. If you want to play around with the data I uploaded one mean and the corresponding stucture as well as the structure after coreg.
http://www.uni-regensburg.de/psychologie-paedagogik-sport/psychologie-kudielka/medien/coreg.rar


Best,

Helmut