Segmentation of Tracts

Hello wonderful MRtrix community!

I was hoping someone had some information regarding segmenting tracts. I have reviewed the previous topic below:

I am wondering if FOD and ACT are necessary in order to segment the tracts. I have generated tracts in MRtrix but my b-value is not powerful enough to run FOD/CSD.

Please let me know. I appreciate your help!

Thank you,
Andrea

Hi Andrea,

So I need to break this one up a little bit:

Tckgen segmentation fault

That post in particular turned out to be in issue with mis-registration between the diffusion and anatomical images, so doesn’t really provide any information with respect to the question you’re asking.

I am wondering if FODs are necessary in order to segment the tracts.

Segmenting tracts relies on the reconstruction of those tracts being accurate; if the tractography algorithm fails to produce streamlines that obey the criteria that you are using in your segmentation, then you won’t be able to segment those tracts as the data simply won’t exist within the reconstruction. There’s been plenty of research in the field that has performed tract segmentation using diffusion models other than CSD, and tracking algorithms other than those provided in MRtrix; we would simply advocate our own methods as we believe they would give the most accurate and trustworthy result.

I am wondering if ACT is necessary in order to segment the tracts.

This depends on the particular mechanism that you intend to use to perform the segmentation. If you intend to use regions-of-interest to select or exclude streamlines based on the streamlines intersecting those ROIs along their length, then ACT is not a prerequisite; you just won’t get the other benefits that ACT provides if you don’t use it. If however you want to segment your tracts of interest based on the locations of streamlines terminations, then I would highly encourage the use of ACT. As with the previous point, there’s plenty of studies in the literature (in fact, probably all of them) that have performed tract segmentation without the use of ACT.

I have generated tracts in MRtrix but my b-value is not powerful enough to run FOD/CSD.

Firstly, I presume from this sentence that you have used MRtrix3’s tckgen to generate streamlines, but using an algorithm that is not based on FODs? If so, then you can simply try segmenting the tracts from those data and see what you find. It’s best to understand streamlines tractography reconstruction and tract segmentation as being two separate stages of processing. In some cases they may inform one another (e.g. tweaking tracking parameters), and in some cases they may be done using a single step (i.e. tckgen using -include / -exclude options), but for most users I’d advise generating a whole-brain reconstruction then using tckedit to select streamlines of interest: that way you get a feel for what’s present in the data, rather than wondering why targeted tracking is not giving you the result you expected.

I would suggest giving CSD a try on your data despite having a lower b-value. You may not get desirable quantitative properties with it at a lower b-value, but it often still performs quite admirably (it certainly doesn’t instantaneously break catastrophically because it detects a low b-value). If you’re falling back to the tensor model instead, I would most definitely recommend giving CSD a go.

Also, a little note on nomenclature: CSD is the algorithm that you apply to the diffusion data; FODs are the mathematical functions that CSD produces (and that you can visualise in mrview). Whenever discussions get a little more complicated it really helps to have everyone on the same page :relaxed:

Cheers
Rob