I now create a binary mask of fixels to keep …
(emphasis mine)
It’s worth noting that the behaviour of the tracking may vary slightly depending on whether you remove fixels that are below some threshold, versus modulating the amplitudes of fixels to ensure that they fall below the global amplitude threshold used within the FACT algorithm in tckgen. Consider a streamline whose orientation is in between two fixels, slightly closer to the smaller one, but the smaller fixel is below the voxel-wise volume fraction threshold. In the former case, the streamline will be attributed to the larger fixel, since it’s the fixel whose orientation is closest to the incoming tangent, and tracking will continue. In the latter, the streamline will be attributed to the smaller fixel, but its amplitude will be detected as sub-threshold, and hence the streamline will be terminated there.
Don’t know how particular you want to be on such details, so just thought I’d raise.
Do you have an idea why the cutoff option doesn’t do what I expect it to do?
It’s hard to know from screenshots alone; a cursory look at the code suggests that the fixel amplitude threshold should be doing as intended. We’ll need to try to duplicate the steps you’ve done here to see if there’s an issue.
I’ve seen that people also use an absolute threshold on the peak FOD …
Firstly, it’s important to be specific about what measure is being used, whether a peak FOD amplitude or a fixel integral. The former is obtained with the -peak option in fod2fixel; the latter is what we prefer to use for FBA, and is what fod2fixel will give you with the -afd option.
This is important because if you want to apply a threshold based on your description here:
3 multiples of the amplitude of a spherical FOD obtained from an isotropic voxel of gray matter
, if you were to derive this amplitude, it would only make sense to apply such a threshold to FOD peak amplitudes, not fixel integrals, and there’s a chance that that may not be consistent with how you’ve been applying fixel amplitude thresholds previously.
If you have a spherical harmonic expansion where the l=0 term is some fixed value (e.g. as averaged from GM voxels) and all other SH terms are zero (i.e. isotropic FOD), then converting that l=0 coefficient to an amplitude requires division by sqrt(4pi).
Rob
Edit: Whoops, that’ll teach me for addressing posts from bottom to top… Already addressed elsewhere.