You could edit the eddy_correct script to use spline interpolation instead of trilinear and that would reduce this effect.
Matt.
Thanks, Jesper!
Follow-up question: eddy_correct introduces a good bit of implicit smoothing, so I'm hesitant to combine non-eddy-corrected datasets (from subjects who didn't move much) with eddy-corrected datasets (from subjects where eddy_correct helps reduce motion). Of course I could "mimic" the implicit smoothing by explicitly smoothing my uncorrected datasets by a small amount. Is that something you'd recommend?
Thanks,
Anna