Pipeline for fornix values (rbx_flow, bst_flow?)


I’ve had my data processed through this GitHub - MASILab/francois_special_spider .
The fornix is my primary interest at the moment. I have a ~18-55 years old population, some with traumatic brain injury, so there could be focal lesions (more frequently they are diffuse lesions though), enlarged ventricles/atrophy. Is it appropriate to use the fornix values as is? In trying to find a precedent (et en lisant la thèse de Dr. Rheault), I discovered that the preferred pipeline may be GitHub - scilus/bst_flow: Bundle Specific Tractography pipeline .
If anyone has thoughts on pros and cons for each/any approach, that would be much appreciated.

Thank you.

Annick Tanguay

1 Like

Hi, Thanks for the great question!

I believe there are two kinds of possible answers.

  1. if you are ok with the possibility of having some subjects with missing fornix or incomplete fornix only because the WM mask had missing voxels (PVE, lesions, etc.), but never be sure why it’s missing, you could keep your data as is. Some people simply use tractography as a way to ‘integrate’ information, so a missing fornix could mean anything, but it means something is wrong. So in this scenario, PVE or lesions lead to reconstruction challenges, which leads to a drop in streamline count or decreased volume, and this is your ‘biomarker’ for something is not going well in this subject. A generic pipeline in this scenario is fine, but the interpretation is a bit more difficult because anything could be the root cause of why the fornix was hard to reconstruct.

  2. if you want to reconstruct something as close as the complete fornix as possible, and you are trying (for example) to map a metric (FA, PET, lesions load) to the fornix, it could be argued that getting a fornix as complete and full as possible is important. In this scenario, where tractography is just a way to get a WM region as close to real anatomy as possible, I believe a pipeline that tries (really hard) to inject priors could be better to maximize reconstruction. And if it fails, then it’s a much bigger ‘red flag’ than in scenario 1) (by that I mean that even custom pipeline fails to reconstruct, it probably means the fornix is not in good shape, rather than simply passing through PVE)

I don’t know your expectations, but both scenarios are acceptable and the goal of BST was to simplify interpretation from a neuroanatomy point of view when we know a bundle is present, but tractography is just inefficient. As for pros/cons, BST does require more processing, but from what you have already done you have the right input for BST saved in the francois_special output. And if scenario 2) sound like what you are going for, I could help you launch BST using a singularity (I think the 1.5.0 version will work with BST-Flow master: Containers)

1 Like

Hi Dr. Rheault,

That’s crystal clear and very interesting.

We are aiming to test the relation between a fornix metric and behaviour, so retaining as many participants as possible is key. I’ll check the rbx_flox outputs with #1 in mind though.

If you could help us with launching BST using a singularity container, that would be very helpful and very appreciated. If it’s easier to communicate directly about this, my email is annick.tanguay@vumc.org .

Thank you.