Trouble while selecting specific tracts from Whole-brain tractogram

Dear MRtrix-Community,

I would be happy if one of you could help me with the following (rather basic) issue:

I would like to extract the transcallosal connection between the left and right M1. Therefore we created a whole-brain tractogram (10mio streamlines, using ACT) from which I tried to extract the tract of interest. For this I used 3 ROIs as inclusion areas in the tckedit command: One ROI was a Corpus Callosum mask, the two other ROIs were build around the peak activation in fMRI to localize M1 for each individual. These ROIs were build the following way: Created a sphere around the peak activation coordinate big enough to intrude the White Matter and then multiplied it with the White Matter mask to restrict it to the White Matter.

Basically this resulted in the desired tracts, but (even though I know it is not a good idea to count streamlines here) I am a bit in doubt if my results are reliable, because:

  1. There are some people with very few streamlines (between only 19 - 80)
  2. There is quite a large variability: Minimum number of streamlines = 19, Maximum = 1.700

My question is: Is this a common result or did I make a mistake or chose the wrong approach?

I would appreciate if one of you could help me!

Best regards, Andrea

Hi Andrea,

It’s very difficult to tell whether or not there is anything fundamentally wrong with your data without being able to see the data itself. Generally the trap that people fall for in this context is using include ROIs that are defined within the GM only, which results in the streamlines terminated using ACT failing to reach them; but it sounds like you’ve got a reasonable mechanism in place to account for this. So beyond that, I can only:

  1. State that reconstruction of streamlines-based connection density has far greater variance than one would hope, due to the inherit ill-posedness of the reconstruction problem;

  2. Suggest that rather than staring at the extracted numbers in dismay, interrogate the data further. E.g. You could try expanding the inclusion ROIs and see whether some of the differences may be due to differences in the number of streamlines just missing your target regions, or maybe generate a mask corresponding to your pathway of interest and then extract streamlines that traverse that pathway for a significant portion of their length but then (erroneously or not) latch on to some other fibre orientation and end up in a different cortical location, and see whether the relative proportions of such trajectories varies in any insightful way.