Weighted mean FA connectomes

I am computing weighted mean FA connectomes from HCP subjects.
I use default commands such as:

tcksample tracks_msmt_5M_01.tck FAmap.nii.gz mean_FA_per_streamline.csv -stat_tck mean 
 tck2connectome tracks_msmt_5M_01.tck nodes_Schaefer2018_200Parcels_17Networks_order+aseg.nii.gz \
				mean_FA_connectome_200.csv -scale_file mean_FA_per_streamline.csv -stat_edge mean -assignment_radial_search 3

and Schaefer 2018 with 200 parcels as parcellation.
The resulting Structural Connectome matrix appears quite “flat” (see figure)

, with very little variations or clusters, especially within hemispheres. Am I doing something wrong? is this normal? I would have expected some more clusters.

Hi Rosella,

A mean FA-weighted connectome is expected to be relatively homogeneous in intensity, especially compared to something like streamline count where edges vary by orders of magnitude. You have a huge regression-to-the-mean effect of averaging values both along streamline trajectories and across the breadth of each bundle. And the FA “fluctuations” within the deep WM aren’t actually all that large. E.g. You can see that many inter-hemispheric connections have larger mean FA values that intra-hemispheric connections as they get the rare privilege of sampling values from single-fibre voxels for a reasonable fraction of their length.

Even if these data were to be “flatter” than other mean-FA-weighted connectomes, the connectome data would provide very little insight into such. If you hypothesize that two edges should be more disparate in mean FA than what you observe, you can try extracting those streamlines using connectome2tck, resampling the underlying FA values, and re-performing those calculations.

Otherwise, the only reasonable a priori hypothesis of what could cause such an effect is if the probabilistic streamlines are being permitted to disperse too much, which results in each edge sampling from a large fraction of the total WM instead of only a small sliver of it. This depends on your tracking parameters, which are not shown.


1 Like

I see. Just one more question…since mean FA does not seem to be a significant metric for myelination degree of the tracts connecting two cerebral areas, which other metric quantyifying this “anisotropy of diffusion” do you suggest?