I have non-isotropic dwi data as 0.9×0.9×3. I would like to do both TBSS and tractography analysis, is it better to resample the image to isotropic voxels?
If so, should I resample the resulting files like FA MD RD AD maps and L1,L2,L3 and V1,V2,V3 ? or should I resample the raw data at the very beginning, like before bet and eddy correct?
I’m not myself aware of any work where the benefits (or lack thereof) of resampling highly anisotropic to isotropic voxels are assessed objectively.
My intuition says:
Denoising and Gibbs ringing can’t be used after any form of reasmpling;
Don’t do any resampling along the slice encoding direction if you intend to perform pre-processing that includes any explicit consideration of slice effects (e.g. dropout, slice-to-volume motion correction)
If you want e.g. tensor parametric maps to be higher resolution, it is preferable to resample the DWI data and then perform the tensor fit, rather than fitting the tensor and then resampling the results; see this work.
I’ve not used TBSS myself, but in the FBA pipeline it’s possible to specify the voxel size of the template image, independently of the resolution of the subject input images; if similar is possible with the TBSS pipeline, that’s a reasonable option.
For many of these processes, it’s not strictly necessary to be exactly isotropic; so e.g. you could instead resample to 0.9 x 0.9 x 1.5, which would make the “effects” of interpolation a bit more consistent across the image FoV (probably doesn’t matter that much, but integer relationships are nice).