 |
 |
Thank you for the quick reply! Yes, that's true of course. But is it possible to define some max values for translations along x y z (say, don't move more than 1 mm)? It is quite easy to correct for these displacements manually, at least if you have enough time. The tricky aspect for manual adjustments seems to be the rotations. So concerning rotations I would rather trust some automatic algorithm. But of course, as you mentioned, due to the dropouts this would be imperfect as well. Alternatively, what happens when setting non-brain tissue and parts with dropouts to 0 or NaN in one of the volumes? Would the coreg be based on the non-NaN voxels then? Or maybe something like coregistration based on the high frequency part of the volumes, maybe even just based on a part of the volume? The shape of the ventricles and the border between cerebellum and cerebrum already contain quite some information. I'm just curious, because the normalisation and segmentation steps have been greatly improved over time. But it is still common to segment the structure and not the EPI. So, what about coregistration? Oh, and it might well be just a theoretical issue, but anyway. I'm just interested :-)