I don't have a specific suggestion, but have you checked the literature for studies that require subjects to speak? You might find some methods for dealing with the motion there. The problem with your design is that the movement coincides precisely with the task. So maybe one of your search terms could be "stimulus correlated motion", and go from there.
On 01/22/2014 10:31 AM, Colin Hawco wrote:
[log in to unmask]" type="cite">
Hello all,

I have a study featuring patients and controls, in which we ask them to imitate an emotional face. Despite trying to train participants to imitate without moving their head, we have quite a bit of motion in the data. Worse, the patients moved more than the controls, representing a potentially very serious confound in our results.

I want to salvage as much of this data as possible, and I am concerned the standard SPM realignment isn't up to the task of dealing with such extreme motion. I recall seeing reference to other realignment algorithms which are specifically designed to deal with large motion in clinical groups. Could someone be so kind as to point out some possibilities to us?

We also plan to use FSL's ICA artifact removal, which I hope will account for some of the artifacts in the data.

If anyone has any additional advice or input for dealing with unusually large amounts of motion in their datasets, I would be grateful for any advice which is given.

Thanks a lot.
Colin