Whole-brain connectome using Schaefer parcellation

Dear experts,

I’m trying to create a whole-brain connectome using Schaefer parcellation instead of Freesurfer atlases.

From freesurfer I have created subject schafer100parcel.mgz file through mri_surf2surf and mri_aparc2aseg. Next, I tried to create the connectome using labelconvert:

labelconvert prova.mgz $FREESURFER_HOME/FreeSurferColorLUT.txt /usr/local/mrtrix3/share/mrtrix3/labelconvert/fs_default.txt subject_1_parcels.mif
Which gives me the connectome in 84x84 parcels.

I think that I need a sort of “fs_Schaefer.txt” file instead of fs_default.txt.
Any suggestion about this process or where I can find the right .txt file

Thanks a lot

Welcome Lorenzo!

If the index values contained within a parcellation image are already incremental from 1 without any gaps, and you don’t wish to apply any kind of custom reordering / exclusion of nodes, then there is no need to apply the labelconvert step, since its core purpose—to produce a parcellation image where the contained values increment from 1 without any gaps—is already satisfied by the originating data. This appears to be the case for the Schaefer parcellation based on the contents of the provided LUT files, and also from my running the labelstats command on one of the parcellation images I chose at random.

If this is not the case if you generate the subject-specific parcellation image through FreeSurfer commands, then the labelconvert step may be required. But as I stated in a prior comment in the originating thread, unless somebody else has created such a file and is willing to share it, I am unsure how to formulate advice for the construction of such a file over and above the contents of the documentation page that would require less effort than just doing it myself.


Hello @rsmith , thanks for your detailed response to this question.

I have a follow up to this:

In a similar case, the atlas i’m using already progresses from 1-268, so the “labelconvert” step was skipped. However, in building the connectome, for some subjects it completed successfully. For others, some nodes were unassigned. I have also looked at some threads for unassigned nodes but not been able to solve it. Please what would you suggest?

Thanks ahead.

Hi Joan,

If you have a parcellation with a comparatively large number of nodes, but that parcellation is mapped to subject space via a volumetric-based transformation rather than a surface-based one, then it’s possible to have parcels in subject space that just due to unique cortical folding there are no streamlines terminating there. Alternatively, it could be indicative of some other problem. This is why the warning message generated in this circumstance simply states that there are multiple possible reasons, and so the only way forward is to check the data closely. Find a parcel with no streamlines assigned, and see where it resides relative to the anatomy.


1 Like

Hello @Joanne

The issue with some of the nodes not being assigned might be related to a too small number of streamlines.

Even with a ~80 region atlas, about 10% of connections have connection strength between 0 and 1. See Fig 4 and Table 1 of my paper:

If I’m not mistaken, about 200 of the connections in my study had exactly 0 strength connection (i.e. no streamlines at all).

With more nodes, you will have even more connections with 0 strength, and it is indeed possible that some nodes might not have any connections at all.

Notice that earlier in the paper, I write:
“As a larger number of regions in these connectomes leads to a larger number of potential inter-areal connections, for this analysis we constructed tractograms with 70 million rather than 10 million streamlines.”

So maybe that might be also a solution in your case.

Good luck!

1 Like

Thanks a lot for your input Rob.

Kind regards,

Thanks a lot for your input Oren.