I have diffusion MRI data at b=1000 s/mm² acquired at two visits. Visit 1 used 64 diffusion-encoding directions, while visit 2 used 128 directions. I’m planning to perform fixel-based analysis (FBA) on this longitudinal dataset.
I’m unsure whether I should:
-
Downsample visit 2 to 64 directions (e.g., by selecting a subset of gradients) to match visit 1, or
-
Keep visit 2 at 128 directions and proceed with the mismatch, assuming that beyond ~64 directions there are no major gains in angular resolution at b=1000.
My main concern is whether the difference in the number of directions could introduce bias in FOD estimation or fixel metrics across visits and whether it’s methodologically sound to keep different direction counts in a longitudinal FBA pipeline.
Has anyone dealt with a similar situation? Would you recommend downsampling to match direction counts, or is it acceptable (or even preferable) to keep the higher number of directions for visit 2 and account for visit/protocol in the statistical model?
Any advice or references would be greatly appreciated.