Evaluating output from tck2connectome

1

Hi all,

Thanks so much for the great software package. I am using MRtrix to construct structural connectomes of stroke patients and have found it very useful so far. In MRtrix, I followed Andrew Jahn’s tutorial (with some editing to tailor it to my own data), and then used the AAL-3 atlas (inverse-normalized to subject space in SPM) to perform tck2connectome. At each step, I used mrview to check the output and it seems to have worked well. However, when I am evaluating the connectivity matrix that is the output of tck2connectome, I’m struggling with a way to evaluate whether the matrix “makes sense.” Any ideas?

Thanks so much!

2

Welcome Celia!

It is indeed difficult to know from the connectome matrix data alone whether or not there has been any sort of issue with connectome construction. I can float some basic ideas:

  • Check that there aren’t any empty rows / columns, which would correspond to parcellation nodes with no streamlines attached. tck2connectome should issue a warning if this happens, but it’s a sensible check nonetheless.

  • Generate a histogram of edge connection strengths. I can’t think of good data to cross-reference it against though: especially with our tracking / quantification the edge weight distribution can be vastly different to that derived using other methods.

  • Depending on what you are quantifying per edge, you could sum the total connectivity of each node, and then see if that correlates with parcel size. For any additive measure, a larger parcel should have “more connectivity”. AAL-3 should have enough parcel size variability for this, but obviously for a parcellation that’s derived based on equal parcel sizing it wouldn’t be of much use.

  • Open up the mrview connectome visualisation tool, and see if the strongest connections coincide with known anatomical connections.

Other ideas welcome!
Rob

3

Thank you so much for your thoughtful response, Rob! From playing around with the connectome visualization tool, the connectome that I’ve generated looks reasonable, and the connectivity (streamline count) does correlate with parcel size. Again, I really appreciate your help, thank you!