TractSeg w/ population wmFOD template

Hello! I’m running into some problem with my tract segmentation masks derived from TractSeg when running it on my population FOD template. I followed the FBA-pipeline as shown by the mrtrix website : Fibre density and cross-section - Multi-tissue CSD — MRtrix 3.0 documentation

One thing to note is that my dataset is all pathological brains (stroke), which I did not explicitly mask out in any steps. The DTI data has 2 unique bvals (b = 0 and b =1000), and I did multi-shell multi tissue CSD, only using the white matter and CSF response functions.

I’m wondering why my tract segmentations seem so far off? I attached the tract segmentations I got for the corpus callosum (entire thing), as well as the left corticospinal tract - they don’t look accurate to me.

This is my first time working with fixel based analyses, so I really appreciate any help! Thanks so much.

Screenshot 2024-10-03 at 2.00.52 PM

Screenshot 2024-10-03 at 2.00.20 PM

I don’t have an answer to your main query, but I am working with a dataset with similar bvals and I noted that doing a single shell multi tissue CSD pipeline provided much better outputs.

You will have to use a specific MRtrix build MRtrix/3Tissue for this → Single-Shell 3-Tissue CSD (SS3T-CSD) | MRtrix3Tissue

You can find a similar discussion about it here → Single-Shell (+B0) 3-Tissue CSD

Thanks for the reply!

When you say better outputs for the single shell CSD, do you mean in terms of the FODs created? Or the actual tract segmentation?

I’ll give it a shot and see how it goes. Cheers!

Hi,

Single shell data worked fine with the multishell pipeline in my own little experience. Regarding the strokes, if no lesions are in the FOD template, I’m not sure it would affect TractSeg’s anatomy-based segmentation substantially either, though I could be wrong.

What I would suggest is rather looking at your TractSeg pipeline and at the 3D tracks it generated. Did you obtain the masks from the latter? Are these fixel masks loaded as overlay in mrview? Plotting tracks or the masks as fixels instead of overlay images could make more sense as you visually get the directions in mrview.

Bundle by bundle, you could simply look at what TractSeg’s corresponding bundles of interest look like and see if that roughly matches their own high resolution example:
https://www.sciencedirect.com/science/article/pii/S1053811918306864

1 Like

I simply meant in terms of FOD. SS3T version is only slightly different from MSMT approach (but you can see more green - csf areas)–> MSMT approach with CSF tissue ditched is very sparse and possibly leads to reduced number of fiber tracks at streamline step. I did a comparison of SS3T vs MSMT (without CSF) vs MSMT (with CSF).

I am still new to TractSeg.

1 Like