Multiple tracts SIFT2


I am doing tractography to retrieve some specific bundles in this way:

  1. get my pathways of interest with tckgen
  2. merge pathways of interests with whole brain
  3. apply SIFT2
  4. extract pathways of interests with tckedit

My question is that since two of my pathways of interest overlap (because one contains the other), if the steps above are still correct. Or should I do these steps separately for each tract of interest? From what I understood from this post, this would not be recommended, right? Is this question of overlapping tracts an issue for SIFT2 or does it not matter at all?

Thank you in advance!

Just to chime in here. My understanding of SIFT2 is that the important thing is that you have the sufficiently-sampled whole brain set of streamlines (e.g. 10 million seems to be a common suggestion). Then, you can add as many more streamlines as you want, including from overlapping ROIs. These additional/overlapping streamlines will just be down-weighted. So, in this case, you would combine all streamlines and run SIFT2 once. You’ll just want to make sure to properly extract and apply the SIFT2 weights in subsequent steps.

Hello Benjamin,

I understand now. Thank you for your reply!

The only reason why merging targeted tractography with a whole-brain tractogram is suggested in some instances is because for very small pathways, the number of raw streamlines in the whole-brain tractogram may be too small for quantification to be reliable, and so the targeted tracking just boosts the raw reconstruction density in the area of interest. Admittedly I’ve not done exhaustive testing on this approach, and there are potential interactions with the regularisation. If your pathways of interest are reasonably large, it might actually be preferable to not use this approach and just rely on the whole-brain tractogram, if only because it’s a reduced level of complexity and less likely to run into issues at peer review.

There is also a deceptive trap here. When you merge tractograms like this, it’s important that following application of SIFT2, you then select all streamlines belonging to the pathway of interest. This is not equivalent to the set of streamlines generated through targeted tracking: some streamlines within the whole-brain tractogram should also be attributed to your pathway of interest. This applies doubly in your case if you have multiple pathways of interest that “overlap”, i.e. some streamlines should be attributed to both.



Thank you for your reply.

I am not sure if I understood what you said here:

You mean that is better to rely on doing targeted tracking only, instead of this approach, since my tracts are large enough and due to the reasons you gave in the last paragraph?

Thank you,

it might actually be preferable to not use this approach and just rely on the whole-brain tractogram

You mean that is better to rely on doing targeted tracking only …

No, performing targeted tracking only precludes the application of SIFT2.

The typical approach is to generate a whole-brain tractogram, apply SIFT2, and then select the streamlines corresponding to your pathway of interest. The optional concatenation of the whole-brain tractogram with the outcomes of targeted tracking prior to running SIFT2 is the component that I’m suggesting may be better avoided unless there is a strong justification for its use; and where it may be justified is in the specific case where once you select your pathway from the whole-brain tractogram, the streamline count is too small. If your pathway is of a reasonable size, and your whole-brain tractogram is generated with an appropriate number of streamlines, this probably isn’t the case, and therefore the added complexity of combining whole-brain tracking with targeted tracking isn’t justified.

Ah ok, I understand!
Thank you for your explanation!