Maybe this particular subject was positioned differently in the scanner, and thus the AC was further away from the origin compared to other volumes. But sometimes it's difficult to understand, some volumes with large replacement work fine and others not. In your case, the algorithm seems to have worked quite well actually (rotations, left-right, anterior-posterior), as far as I can see it only failed along the ventral-dorsal axis ;-)
BTW, similar problems can arise during coregistration and normalization.
Best,
Gabor
|