I have performed nearly all preprocessing steps, but I’m not sure why the analysis has to be performed in template space. I am interested in the fiber density and cross section of the individual corticospinal tract (as identified with CSD tractography) and how this changes over time. I feel like I am losing subject specific information when I identify the corticospinal tract in the fod template, especially because different subjects have different lesion locations and sizes.
So would it be possible for me to calculate the FD of the individual corticospinal tract in subject space? To obtain information on the cross section of the fiber bundles I could calculate this from the template warp and transform this back to the subject space.
Analysis needs to be done in template space in order for individual fixels to be comparable across subjects. If you are to instead collapse all information from the CST into a single quantity per subject (per metric of interest), then technically yes this can be done in subject space rather than template space, on the understanding that the number and location of fixels used to derive those quantities will almost certainly vary between subjects.
The question then becomes: how does one delineate the CST in subject space? Because another advantage of working in population template space is that if you want a fixel mask corresponding to the CST, you only need to define one such mask, and it is then applicable to data from all subjects; whereas obtaining good fixel masks in individual spaces in a fully automated fashion may be difficult.
Once you have such masks, it’s just a matter of running mrstats on an individual-space fixel FD data file, providing that mask.
If you’re interested in achieving this, I find that it can be done very robustly with TractSeg (GitHub - MIC-DKFZ/TractSeg: Automatic White Matter Bundle Segmentation) these days, which is directly compatible with MRtrix’s FODs or peaks. In case of certain pathologies, they can end up being more (better) subject-specific delineations.
Note that averaging over fixels was typically not called “fixel-based analysis” in the past; but technically, it of course still is (since it’s literally fixel-based, disentangling contributions from other crossing structures). It’s just not fixel-wise analysis (i.e. per individual fixel, matching across subjects). But that’s all just semantics of course.
So in conclusion, your general thinking here is in fact very reasonable. Take a look at the tools that come with TractSeg for sure!
Following up on this, is it possible to estimate fiber cross-section in subject-space, since you don’t have the warp field from the template registration?