Dear Colin,
You could still check whether co-registering bias-corrected EPI means and bias-corrected, skull-stripped anatomies leads to more consistent results across subjects (in case you haven't done so already), although this would be a rather general "solution".
In principal you could also go with skull-stripped bias-corrected EPI means, with which you could ensure that the EPI images only contain brain voxels, but it might be tricky to find a good threshold for the skull-stripping due to rather large voxels and partial volume effects.
Accordingly, possibly it is more promising to turn to a different algorithm, see e.g. Saad et al. (2009, http://dx.doi.org/10.1016/j.neuroimage.2008.09.037 ) or Greve and Fischl (2009, http://dx.doi.org/10.1016/j.neuroimage.2009.06.060 ) for some not very new but established ones. Another nice routine is the "Point alignment" option ( https://web.stanford.edu/group/vista/cgi-bin/wiki/index.php/RxAlign#Fine_manual_alignment ) available in mrVista. Basically you can select prominent landmarks in both anatomy and EPI as a starting point, after a few iterations you select new or additional landmark and so on, thus you can optimize coregistration for certain regions. Whether this is really necessary in your case is another issue, but it's very useful for e.g. optimizing registration for occipital lobe in case you want to do retinotopy analyses, and it might always be worth a try in case one runs into problems otherwise. Last but not least, there are quite many softwares or tools that deal with between-modality registrations specifically, although not necessarily optimized with regard to T1-EPI.
Best
Helmut
|