The question is probably very basic, but it suddenly made me stumble. I always used a structural image as reference, and mean (or first) EPI image as source image, assuming SPM will wiggle mean and the rest of the EPIs to match structural which is then normalized. So far it seems to be working just fine. But now that I've taken a closer look, SPM 12 manual has it vice versa - have mean functional image as reference, and wiggle the structural image as source image. Is it a better approach than the one I've been using? If so, why? If I have 9 runs, and I realign each run separately, I will have 9 means, which means I will have 9 different structural images. Is it a better approach to normalize it this way?

Sincerely,
Natalia Yakunina