error with streamline assignments

Dear experts,

I am performing tractography with mrtrix3 v3.0.1 and using a custom parcellation to get a connectome using tck2connectome. When I check my results using connectome2tck with streamline assignments from tck2connectome, i get noisy fibers in other parts of the brain (that don’t start or terminate in either node I’m looking at). I haven’t reproduced this error yet (by repeating the tck2connectome and connectome2tck runs) but I was hoping to see if anyone else has had this error because it seems kinda serious. I’ve followed the multi-shell pipeline (I’m using multi-shell data) with denoise, topup, eddy, bias correction and finally generating 25 million streamlines using tckgen.

these are the commands I’m running:
tckgen wm.mif 25M.tck -act 5TT.nii -backtrack -crop_at_gmwmi -seed_dynamic wm.mif -minlength 10 -maxlength 250 -select 25M -mask brainmask.nii -output_seeds succeeds.txt -force -nthreads 8

tcksift2 ${numfibers}.tck wm.mif sift_weightfactor.txt -act r5TT.nii -fd_scale_gm -force -nthreads 12

tck2connectome 25M.tck parc.nii parc_streamweights.csv -tck_weights_in sift_weightfactor.txt -assignment_radial_search 2 -zero_diagonal -out_assignments streamlineassignment.txt -force -nthreads 12

connectome2tck 25M.tck streamlineassignment.txt subj -nthreads 12

This is the error that I’m getting:

As you can see, I get the streamlines from the nodes at the front left (right on image) but I also get tracts from the back of the brain as well that don’t seem to come from anywhere.

There’s nothing wrong with the tractography (that I can tell):

And here’s my parcellation:

The nodes are custom masks informed from fmri. They are discontiguous because parts of the nodes are in WM (so I cut those out) but I don’t see why this should be causing an issue.

This error is common among subjects as well.

I would appreciate some advice here on how to proceed; whether I just use thresholding to get rid of the noisy fibers, or what the cause of the error might be (if it is indeed an error?)

Nikitas Koussis

Welcome Nikitas!

With the nature of your report, there is such a wide range of possibilities on the table that I’d be taking wild shots trying to diagnose. If you are able to make example data available to me, it would likely save me a huge amount of time compared to trying to walk you through what I would otherwise be checking manually myself. But I would definitely advise that the data be checked closely rather than trying to hide a potential issue through thresholding.