I read the very helpful page under “Concepts” and “Motivation for afdconnectivity”, and related discussion posts. I wanted to clarify - if I have diffusion data for 2 timepoints (before and after a surgical intervention), but unfortunately no EPI distortion correction images, and wanted to quantify changes in a few particular pathways, what is the best method? My understanding is I could either:
- Follow the standard preprocessing pipeline, generate a whole brain tractogram for each subject/each timepoint, apply SIFT, tckedit to get specific tracts, and then compare streamline counts.
- Follow the standard preprocessing pipeline, generate whole brain tractogram for each subject/each timepoint, tckedit to get tracts of interest, then use afdconnectivity with the -wbft
I’m wondering if these two approaches are correct, and if one is much more valid than the other?
Thanks a lot for your help!
As far as quantifying the total fibre intra-cellular cross-sectional area of particular pathways of interest, both approaches are “correct”, albeit with different assumptions / physical constraints / pragmatic issues. I have a draft manuscript describing the differences here - which I should have published ages ago - that would otherwise have formed a good demonstration of such; but alas. While the SIFT approach is more faithful to the physical reality of white matter fibres, both approaches suffer from problems related to what I now refer to as “attribution”: selecting which streamlines belong to the pathway of interest and which do not, which becomes highly ill-posed in the absence of something like ACT.
The other option is to follow more of an FBA analysis pipeline, but instead of performing statistical inference at the fixel level, instead generate fixel masks in template space corresponding to your pathways of interest, calculate for each subject the mean value of some fixel-wise quantitative parameter (e.g. FD) within that mask, and then perform statistical inference using those summary statistics (one scalar value per pathway per subject). In the absence of EPI distortion correction, which makes subject-wise processing very difficult, this may actually be the more robust option, even if the derived quantity is not precisely identical in physical units or interpretation to “pathway connectivity”: e.g. in the case of mean FD within a fixel mask, what you’re actually quantifying is something more like a “mean fibre volume fraction within pathway” rather than a “total fibre connectivity of pathway”. FDC is perhaps better in that respect, but the interpretation is still convoluted.