Challenges with Seed Generation using '5ttgen fsl' for Anatomically Constrained Tractography

Hello MRtrix3 Community,

I am currently working on creating structural connectomes using anatomically constrained tractography.
For this purpose, I have been utilizing ‘5ttgen fsl’ to generate tractography seeds.
However, I’ve encountered a significant issue: in about 10% of my dataset, which includes almost 100 subjects, the seeds do not seem appropriate, specifically, in the parietal region of the brain.
These are examples of a poor seed and tractography.


I suspect this problem may be due to the poor quality of white and gray matter segmentation.
(And I guess that could be partly due to inhomogeneities in the T1 signal intensity.)
These are part of the result of segmentation(GM and WM).

While one potential solution is to redo the process using ‘5ttgen freesurfer’ from the beginning, this approach seems labor-intensive.
Additionally, it feels wasteful to discard the successful results in the remaining about 90% of cases.

I am reaching out to ask if there are any preprocessing strategies or other techniques that could improve seed generation using ‘5ttgen fsl’. Any insights or suggestions to address this challenge would be greatly appreciated.

Thank you in advance!

Best regards,


Hi Kenbridge6,

Running bias field correction on the anatomical image before 5ttgen fsl may help produce segmentations that are more accurate… I cannot guarantee that it will improve the segmentation quality, but might be worth a go :slight_smile:



Hello, I had the same problem with some of my data. It’s unfortunately caused by the acquisition itself… scanning in axial orientation with slab-selective exciation produces low signal regions near the upper and lower edges of the slab… unfortunately, I didn’t find a way to eliminate this in post-processing, but the above suggestion may improve the results.

1 Like

Thank you for your advices!
I’ll try bias field correction before ‘5ttgen fsl’ and see the results.