No streamlines between pairs of ROIs in normative connectome (Elias et al. 2024) — single-ROI include works, dual-ROI include returns zero

1

Hi everyone,

I’m exploring connectivity between three ROIs — dlPFC, vmPFC, and amygdala — using the large normative connectome from Elias et al., Scientific Data 2024 (dTOR_full_tractogram). I converted the published .trk to .tck with DIPY (world coords, RASMM). Visualization in mrview looks correct in MNI.

Including one ROI with tckedit returns many streamlines, e.g.:

tckedit dTOR_full_tractogram.tck dlpfc_left_filtered.tck \
  -include dlpfc_left_1mm.nii.gz

Including two ROIs together (e.g. dlPFC-L and amygdala-L) returns zero streamlines:

tckedit dTOR_full_tractogram.tck left_dlpfc_amyg.tck \
  -include dlpfc_left_1mm.nii.gz \
  -include Amyg_left_1mm.nii.gz

My question:
Is this expected behavior, should I use a different tractogram, or have I likely made a mistake in my setup?

Thanks a lot for your help!

2

Welcome @biederm!

I wonder if anyone in the community would be willing to take on a “tips and tricks for targeted tractography / track editing” Wiki page. I’m sure such a resource would be of considerable utility in the domain. It is however not something I regularly engage in myself, so while I can give some suggestions, they are based on the underlying conceptual framework and not real-life experience.

  1. The tractogram described is based on deterministic tractography. This technique is known for its vulnerability to false negatives. While there should be a lot of variation in streamline trajectories across individuals, maybe that isn’t enough to capture trajectories that are plausible but non-dominant.

  2. With targeted tractography / track editing, it seems people regularly get trapped in the “run command → no streamlines → hands in air” routine. I would encourage thinking about how you might go about interrogating the data. For instance, given here you are not generating streamlines but using an immutable tractogram, you could try dilating your masks and re-running tckedit, to see if there are any streamlines that are close to being ascribed to your pathway of interest but don’t quite touch the ROIs.

  3. If either individual participant or template image data are available, you could try running probabilistic streamlines tractography.

Regards
Rob