Hello again FSL experts!
I'm writing to ask what I assume is a quite trivial question, but I really want to be sure before investing to much on what could turn out to be the wrong solution. I scrolled through the jiscmail but didn't find any satisfactory answer comparable to my problem.
So I have a T1 at 1x1x1mm, an active-fmri at 3x3x4 and a RS fmri at 3.5x3.5x4; I'm interested at comparing certain ICA components versus task-fMRI.
The thing here is that I was doing registration manually (using Nudge) and applying the matrix to the thresh_zstats from the melodic output, after mixture modelling. My images looked a little strange and I eventually came across the problem of interpolation. Basically, since I have different image sizes, when I FLIRT the zstats to the T1 image I get a lot of interpolated values below my threshold because the data is obviously not continuous. I don't have a good intuition on what Nearest Neighborhood does in cases of anisotropic voxel sizes. So my question is: can I just FLIRT both task and resting state fmri thresh_zstats to t1 using NN and I wont get problems? Another solution is to change the pipeline in a way I actually run the registration on the motion-corrected EPI data (upsampling a-fMRI and RS-fMRI to 2x2x2 and downsampling T1 to 2x2x2 for computing efficiency), and only then run the analysis, but I'm a little concerned this might have implications on the ICA decomposition. I'm also planning to use Sørensen–Dice index so registration is really key to get accurate results; Could you enlighten me about this issue?
Thanks once again,
PB
|