Increased FA value in a non-crossing fiber region

Dear MRtrixers,

I’m studying a neurodegenerative disease using the FBA framework and trying to compare the results to traditional voxel-based DTI (in a similar fashion to Mito et al., 2018).

Part of my analysis is to study the correlation between a specific clinical measure and degenerative WM changes (using both FBA and VBA). The lower this clinical measure is, the more WM degeneration is expected. FBA correlation results show exactly that, while VBA showed the same relationship but not in all tested tracts.

While I was checking both positive and negative GLM contrast results, I noticed a cluster of around 4 voxels showing a statistically significant increase in FA correlating with the decrease in the clinical measure (opposite of what’s expected).

At first I thought this could be a similar effect to what Mito et al. found with areas of crossing fibers, were an increase in FA was found in the AD group and was attributed to the tensor model failing to model crossing fibers. However, I checked the fixels in the region with the increased FA (optic radiation near the occipital pole) and couldn’t find any crossing fibers. Plus, as far as I know, this region isn’t particularly known for having crossing fibers.

Now I’m trying to find an explanation for this small area of increased FA. Could there be crossing fibers that area that was not accounted for in the fixels, due to perhaps some limitation in the data? Any idea as to why I’m getting this increase in FA?

Looking forward to hear any insight on the matter. And as always, thanks for the great software and the constant online support!


Hi Joe,

A few ideas:

  • 4 voxels is very few… never exclude the chance that… it happened by chance.
  • I’m not entirely sure that area is positively “free” of crossing fibres. Also with CSD, especially when using thresholds to extract discrete fixels, we’re not 100% sensitive to everything.
  • Bending fibres, although less likely to cause this, but it’s not impossible.
  • …and last, but certainly not least: other tissues. In that area, and depending on your voxel size, I wouldn’t be surprised to see GM or GM-like signal contributions. If those would’ve changed (in terms of degeneration), there’s scenarios where that can increase FA as well.

All of that just to conclude there’s (even) more issue with FA than merely the crossing fibre behaviour. Some of those problems are removed with CSD. Some more are removed with full 3-tissue CSD. :wink:


1 Like

Hi Thijs,

Thank you for the prompt reply! I also considered the effect of other tissue compartments, especially since the voxels are bordering the occipital cortex GM. But thank you so much for the other possible explanations!

Can I get your personal opinion on something? Do you think it’s a good idea to make an argument against the tensor model (vs. CSD/FBA) using this finding? Just as a side note, not a major finding or anything. Or would it sound ridiculous, given the small number of significant voxels (5 to be exact :sweat_smile:)?

Thanks again!!


I wouldn’t say it’s ridiculous, since there’s a sensible technical explanation for sure; but I agree that such a small number of voxels is hard to drive the argument that you’re actually witnessing this in your data at hand. So definitely a side note at most, and even then not really as an “argument” against the tensor model probably, but just a curious observation (that has the above as a possible explanation). Does that make sense? :slightly_smiling_face:

1 Like

Yes, makes perfect sense. Thanks for the help :smiley: