Hi Romain,
With back-tracking I’d be concerned about a factor over 2 or 3. You definitely want to compare the output of mrinfo of those problematic subjects against the rest of the group to see if there’s anything glaringly obvious.
It also looks like you’re tracking on upsampled data without explicitly setting the maximum streamline length, hence ~10% of streamlines are exceeding that length and being rejected. The default maximum length is 100 voxels, which for upsampled data can be less than the length of the white matter.
A few other things you can check:
- Look at the number of directions being used for rejection sampling calibration between the different subjects.
- You haven’t used MSMT CSD? This may introduce issues if used in conjunction with ACT when using the default FOD amplitude threshold.
- Extract the l=0 term of the FOD image of each subject and compare across subjects. If you’ve got gross differences in either DWI intensity or response function size, this could result in inconsistent scaling of FOD size across subjects, which will influence streamlines termination.
- It may not help a great deal here, but it’s a neat trick nonetheless: If you remove the double-slash at the start of this line, and recompile, then
tckgenwill create a set of images in the working directory, highlighting the spatial distribution of streamlines terminations under different termination mechanisms.
Rob