Brainstem / FreeSurfer / Connectome


I’m just starting using MRTrix; and working to optimize the pipeline for our research needs: we are interested in the brainstem. Our MR acquisition protocol is optimized for a very detailed brainstem imaging, and my global tracts and FOD done with MRTrix look rather promising, however, I am concerned about the connectome. I know that the general use of FreeSurfer omits the brainstem segmentation. The nodes generation is based off the FreeSurfer segmentation, so am I right to expect that the quality / details of connections around the brainstem to be lacking? What would be the best (and most time efficient) way to go about it?

Thank you!

The brainstem is indeed a really interesting, but challenging area. Connectomic analysis of this region is going to be difficult, since as you say there are no segmentations of the brainstem available. The main problem there is that there is no MR modality that provides any contrast that would be appropriate to distinguish the various brainstem nuclei. It is simply not possible to segment them using MR images (at least not to my knowledge). If you were to obtain a segmentation of the brainstem, with current approaches I think it would by necessity be a simple transfer of a predefined atlas onto the subject space, with no information to guide the alignment within the brainstem itself (it would basically be a best-guess of where these nuclei might be, with no scope for subject-specific refinement).

But even if you were to achieve such a segmentation, there is another problem: most brainstem nuclei allow fibres to pass through without synapsing (e.g. the cortico-spinal tracts will pass right through the ponto-cerebellar nuclei), but some fibres will terminate and synapse. Unfortunately you have no way of knowing during the tractography process (or even afterwards) which streamline will synapse and which will go straight through. It’s not possible to apply a robust termination criterion for the streamlines - the best you can probably do is use a statistical approach (i.e. if you know that 50% of the streamlines will terminate and synapse, you can impose this constraint to the overall tractogram, but there’s no way to know which of the incoming streamlines should terminate). So even assuming your segmentation is accurate, the connectivity values you get will still have some degree of arbitrariness.

That said, the brainstem is undeniably a fascinating region to study - I’m just trying to point out some of the more obvious difficulties in doing so well. Ultimately, I think anything you do will necessarily involve the application of a lot of prior knowledge to somehow ensure plausible results. The problem in these situations is that there comes a point where the technique you use essentially adds nothing to the knowledge that was imposed to constrain the problem in the first place…

Good luck with your investigations, and I hope you find a way to deal with these issues!

Better late than never! […re-visiting this question now, realized that I responded in my head] THANK YOU!

1 Like