I’ve run the pipeline to estimate FC, FD, and FDC of my single-shell DWI data. Then, I decided to use ROIs (tracts including the uncinate, IFOF, ILF, SLF, corpus callosum) to extract mean FBA metrics per tract. I converted these binarized FSL ROIs to fixel space via voxel2fixel, where the output is a directory containing index, directions, and a [tract].mif file. When I view the output [tract].mif file (thresholded at 1), it looks to be in the right location. I did not threshold the mask after or during the voxel2fixel step since this mask was binarized beforehand. Then, as suggested in previous posts, I used the following command:
mrstats -mask [tract].mif -output mean [subj].mif.
Seems good so far? However, when I looked at correlations between tracts, they seemed unusually high. For example, in my sample, the correlation between FC of the left corticospinal tract and right anterior thalamic radiation is .97. Other correlations ranged from .69-.96. Do we expect FBA metrics to be highly correlated across tracts, where greater/lower FC in one tract is likely to be accompanied by greater/lower FC in other tracts? Or, could it be that my method of masking and extracting values is missing something?
I converted these binarized FSL ROIs to fixel space via voxel2fixel
Note that for any voxels within those masks that contain more than one fixel, all fixels within that voxel will be included in such a mask. This is not ideal; typically such information is combined with template-space tractography to try to select only the fixel of interest. Admittedly the suite of tools for performing such is slightly lacking at the moment though…
For example, in my sample, the correlation between FC of the left corticospinal tract and right anterior thalamic radiation is .97.
Does your analysis include total brain volume as a nuisance regressor? If not, then I would expect reasonably high correlations in specifically FC between different bundles.
For FD, it is worth bearing in mind that the various processes for global intensity normalisation will affect all bundles within a subject equally. So this would be expected to elevate correlations in FD measures between bundles, with the magnitude of that elevation being proportional to the errors / biases in intensity normalisation. E.g. Imagine taking the FD images of three subjects, globally multiplying subject 1 by 0.5 and subject 3 by 2.0, and then looking for correlations in bundle A mean FD and bundle B mean FD across the three participants: the botched intensity normalisation with swamp any genuine biological variability. Including the global intensity multiplier applied by mtnormalise as a nuisance regressor might reduce this effect for the sake of this specific type of correlation.