Dear Natalia,
> the position of the head between the sessions may change a little more (up to 3-4 mm
I would be very surprised if Realign couldn't handle between-session displacements like that. In fact, although some geometric deformations do occur depending on exact position, a volume from the second session might be more similar to a volume from the first session if the displacement amounts to exactly one voxel than to another volume of the same or different session with a displacement in the range of half a voxel.
> but then both sessions are realigned to the first image of the first session, and as a result, all images of the second session have to undergo substantial realignment
Why not simply realign the two sessions onto the mean image of the EPI series? It's sufficient to reslice the mean EPI image, you don't have to reslice the EPIs themselves. The necessary reorientation parameters are stored in the header part of the fuctional files and are taken into account during normalisation then, so it shouldn't matter whether one image is jiggled to a larger extent during Realign: Estimate & Reslice, as they undergo interpolation anyway.
> I'm afraid it may also affect the preciseness of coregistration.
In general I feel safer when realigning within modalities first. In principle, if you have doubts about quality of the realignment step then you should probably turn to a mere between-modality registration, i.e. reorient every single EPI image onto the anatomical scan (which would be very uncommon, but if you think it out it might be the only reasonable conclusion then).
Best
Helmut
|