I’m analysing diffusion MRI data and would like to look at tract-averages of FA and MD. I’m trying to choose a tract segmentation method and at the moment it’s between TractSeg and Tracula, and the option of just applying an atlas to mask the images. The data has one b0 image and 60 directions of b1000. Is any method particularly well or badly suited for this kind of data, or is it a question of preference? Is TractSeg overkill with only one (or technically 2) shell(s) and no CSD performed for analysis? On my multi-shell dataset I feel like TractSeg makes more sense since I did CSD anyway for analysis and could use the peaks for both purposes, but with this other dataset I’m not sure. Thank you in advance for any insights!
I wouldn’t say there is necessarily a single best method, and you can always try multiple pipelines and see which bundles look best. If you have preprocessed data from QSIPrep, QSIRecon directly supports several pipelines, including full PyAFQ, MTtrix → PyAFQ, and DSIStudio AutoTrack. An advantage of this is that it provides outputs in BIDS-valid derivative directories. TractSeg and TRACULA as far as I can tell both have stricter (and not BIDS valid) naming conventions. There is a TRACULA BIDS app but I am not sure how well it has been maintained.
You will have to be deliberate in how you average microstructure within the bundle because while I imagine all methods will be approximately equal in capturing the core of each bundle, individual branches and end points may differ between methods. Making sure to use a some kind of weighted average (as opposed to a simple binary tract mask of “if there is a streamline in this voxel”) will help mitigate effects against implausible branches or extensions into gray matter, if that does happen for particular pipelines. I suppose an advantage of PyAFQ is that it will, as part of its pipeline, do the along-the-tract statistics as well. Ad advantage of DSIStudio AutoTrack is that it will also return tract macrostructural metrics (e.g., volume, surface area) that may be interesting to look at too.
TractSeg will require CSD, and I recommend SS3T to do this if you choose this route.