I think empiricism would be the way to find out what works best for your data. I'd guess that if you can achieve a nice segmentation of the EPI, then you are probably better off estimating the alignment based on these data. What works best is likely to depend quite a lot on whether your EPI are spatially distorted. If the are distorted, then (rigid-body) aligning them with a structural scan is unlikely to work well.

Best regards,

-John

On 16 September 2015 at 13:09, Clas Linnman <[log in to unmask]> wrote:

Hi,

I am working on a 7T dataset where the functional data (1x1x1mm) is crisp enough to be segmented, especially the realigned mean functional. The EPI and structural segmentations look slightly different, so my question is which one to use as input for creating a Dartel template? My thinking is that by using the EPI, I can avoid coregistration errors. Or should I use the structural (even crisper), or a combination of the two?

with my best regards
Clas