Hi,
I would go for the 2x2x2 option as it is isotropic and it better represents overlaps in the transformed images (as they don't general cover whole voxels exactly). The increase of the number of voxels is taken into account with the statistics as they all have some form of explicit or implicit correction for the number of independent measurements (which is unaffected by the number of voxels).
All the best,
Mark
On 14 Feb 2013, at 20:31, Ricky Savjani <[log in to unmask]> wrote:
> Hi FSL Wizards,
>
> I had a quick question about registration. For my level 1 analysis (within-subject COPE generation), I plan to keep my 4D epi data in subject space (no registration). I have generated the transforms of the mean EPI after realignment but before smoothing to first the subject T1 and then to the standard template, keeping both transforms. I actually used ANTS to perform both affine and diffeomorphic transformations for both steps. The registration works well.
>
> Now, I am preparing to perform my inter-subject level-2 design (using FLAMEO). I can apply the transforms to the generated COPES and VARCOPES to align all of them to the template. I am wondering if it will be best to apply the transforms and generate them with voxel size (2x2x2) and FOV (91x109x91) of the template space, or to keep them in the voxel size (3.4x3.4x4.0) and FOV (64x64x37) of the functional?
>
> Is there any reason to prefer one FOV and voxel-size vs. the other. The downside of using the template might be that I increase the # of voxels overall, which might have penalties for multiple comparisons corrections?
>
> Thanks in advance for the help!
>
> -Ricky
|