AFD and FBA help!


#1

Hello,

I have rat brain DTI data from mild and severe TBI models (1b0 + 60 directions with b3000 at 0.150 mm^3) for which I would like to perform AFD & FBA.

Based on the previous posts, it seems that it would be beneficial to use msmt_csd, dwi2fod with only the WM & CSF responses, and finally proceed with the multi-tissue FBA pipeline.

The issue is the severe TBI brains have large deformations resulting in template building/registration issues :confused:

So, my idea was to compute an AFD scalar map, and do an ROI based analysis (since whole brain VBA seems unfeasible…)

Does this make approach make any sense? I’m looking forward to your comments :slight_smile:

Karthik


#2

Makes sense, but (as I mentioned in my email): still be mindful of the response function selection, and whether it has gone well, so the response functions make sense; since the result of your CSD outcome and its interpretation depend crucially on it. I’ll add another reply to your other forum post about that, briefly mentioning some details I also emailed, for future reference for people that may be looking into analysing challenging data.

Yeah, that sucks. :slightly_frowning_face: …well, especially with a voxel- or fixel-wise analysis in mind, because everything has to align if you want to compare like with like. If this fails, most studies would just suffer of increased variance, without the risk of false positive findings (if anything more likely to lose results due to lack of power). However, since these deformations are in this case specific to one group, there is effectively a risk of detecting false positive differences that have nothing to do with the pathology (but with the consistent misalignment).

Makes sense again, but if you want, you can make this more specific by making that not just a voxel-ROI using the WM FD scalar values (“total voxel-wise FD”), but effectively a fixel-ROI. You can first use tractography on your template itself to find the bundles of interest. The next step would be tck2fixel, which essentially generates a fixel-wise track density image (i.e. equivalent of track count, but counted per specific fixel rather than just per voxel). Using mrthreshold or mrcalc you can then threshold the fixel folder’s datafile that resulting from tck2fixel, which would leave you with a binary fixel-wise mask (the equivalent of what a ROI/mask is in voxel space, but then for fixels). And then extract FD values only for these fixels.

This would essentially be slightly more specific (and hence potentially sensitive to just the bundle of interest). But if you’ve got success with just voxel-ROIs in the way you describe, that would be useful as well of course. So in conclusion: yep, the approach you suggest sounds like it makes sense. As always, take great care at each individual step, especially in a challenging case like yours. :wink:


Tck2fixel: streamline count or streamline density?
#3

Dear This,

Thanks a lot for the answers and suggestions :slight_smile:

I’m first trying to get the response function estimation working neatly (based on the previous thread & emails :))

For the severe TBI brains, the fixel based ROI measures can be very interesting (vs ROI based measures).

I will work on them eventually (after analyzing the mild TBI brains ;)) and get back to you when I have some results.

Thanks,
Karthik