Number of Tracks

Hello Mrtrix community,

I am working on generating the corticospinal tracts using Mrtrix3 software.
I’ve been using -number 7500 when generating the tracts, but I am wondering if there’s a rule-of-thumb that most people are using for the CSTs. I find that I get several spurious fibres with 7500, including corpus callosum and cerebellum, and sometimes fornix.

Would be really happy to discuss this!

Please and thank you!

Hi @akuczynski,

It’s been a while and no answer, so I though I’d chip in. I don’t think there’s any hard rules about how to delineate the CST (or any other tract for that matter). There may be guidelines, but there’s nothing stopping you from trying other approaches of you think they work better.

The number of streamlines is probably the least important thing to worry about, as long as you have sufficient numbers to get a solid representation of the tract. For most tracts, you will get false positives - or at least stay streamlines that might belong to a different tract. This is exacerbated to some extent using probabilistic tracking, and is definitely made much worse using low quality data (low SNR, low number of DW directions, low b-value, etc.). But the point is, these aberrant streamlines provide information about the uncertainty of the tracking process. The more of these you have, the less certain you can be about where these streamlines should go.

This is why ROI placement is so important. This is how you inform the algorithm of your prior knowledge and assumptions. It’s unfortunately the case that tractography is rife with false positive connections, so anything you can do to avoid them in the first place is good. The skill is in ensuring that these ROIs are informative to the algorithm, yet not so restrictive as to make the tractography process redundant (e.g. putting waypoints all down the tract so there is no alternative but to delineate what you’ve already drawn…). And importantly, these decisions need to defensible and informed by the literature or some explicit assumptions - depends what the question is.

So to get back to the CST: all I can suggest is to try different things and see what works. You will get the odd errand streamline, and you’ll need to make sure your processing can deal with this. I recommend you play around with seed placement as well. For instance, I find that tracking from the motor strip down to the spine using the -unidirectional option provides the most complete of the whole homunculus - but delineating the motor strip can be laborious (something you might be able to get around using something like a template-delineated motor strip registered to native subject space).

OK, all from me. Not sure that’s necessarily all that helpful, but It might prompt others to chime in with their own recommendations…?

Hi Andrea,

Perhaps a more concise point to add to Donald’s comment: you don’t want to rely on reducing the number of tracks to reduce the prevalence of spurious fibres. This should be handled appropriately using your regions of interest (both include and exclude), such that spurious fibres are eliminated regardless of the number of streamlines you generate. The number of streamlines you choose to generate should then be focused on reconstruction of your pathway of interest, rather than the absence of false positives.

My general rule-of-thumb is: “if you can see individual streamlines, you don’t have enough streamlines”. Each streamline is a discrete sample within a continuous field of connectivity; therefore you need enough streamlines for the continuity of that field to be apparent. You also need enough streamlines for test-retest reproducibility to not be significantly affected by repeating the tracking experiment on the same image data.

Hopefully that all gives you a starting point

Thank you very much @rsmith and @jdtournier! I am very bad at checking back the community page for responses to my comments, but I really appreciate your feedback! (And thank you both for your comments on my previous inquiries).