FOD-based WM seed_image mask for whole-brain tractography (and avoid areas below -cutoff)

I have some technical reasons for wanting to seed uniformly throughout WM without seed_dynamic, but I don’t want to rely on freesurfer, FSL FAST, or DTI FA to define that mask due to questions about accuracy. Is there some robust way to determine a WM mask after fitting FOD but before tractography? The first volume from the dwi2fod wm output (is that the 0th harmonic?) looks very white-matterish, but I’m not clear on what exactly that represents. I want to avoid placing seeds in locations where they will be immediately terminated by my -cutoff 0.05 argument, but I can’t tell how to make such a mask from the input data.

The first volume of the WM for image is indeed the overall WM density per voxel, so that sounds like exactly what you’re after. You’ll probably need to scale it up or threshold it at some level, as seeds will be generated in locations when the intensity of the seed image is greater than 0.5.

Thanks! Is (fod_vol0>=0.05) exactly equivalent to the mask I get in the final tractography output with -cutoff 0.05?

No, since that cutoff applies to the FOD amplitude, not the mean FOD for that voxel. In other words, the cutoff used during tracking is direction-dependent. Tracking will stop when the algorithm can’t find a direction to follow that is both within the range of allowable directions and where the FOD amplitude is above the cutoff. You can in general find voxels where the mean FOD amplitude is below cutoff, but the peak FOD amplitude is above cutoff and will therefore be a candidate for tracking.

And it gets more complicated still for iFOD2 since that cutoff applies to the average amplitude of the FOD along the curve segment projecting out from the current location, which will in general span half a voxel or so – not a point location. So it’s both direction-dependent and distributed in space…

If you need to ensure termination at precisely the same location, using the same image for both the -mask and -seed_image should do the trick.

Got it. Is it fair to say the first volume of the WM image is better suited than FA for identifying voxels that are white matter but have crossing fibers?

Yes, I’d be happy to stand by that statement.