Hi MRtrix3 forum!
I am now in front of a task to performe statistical analysis for “Single-case control” study in the best possible way.
For my case, I need to compare 1 subject’s fd, fdc and fc to 100 controls.
As far as I understand,
fixelcfestats perform non-parametric one-sided permutation testing. The permutation part here is what I am concerned about - in my case, this would mean that not 5000, but only 100 permutations will be applied, as shuffling 100 healthy controls - that is not a sufficient number for permutation statistical test.
However, I really like the idea of FWE correction using CFE and would like to use it.
One thing that came to my mind was using other statistical approach on fd, fc and fdc values vectors - not a permutation one. Maybe t-test (teeny-tiny ) slightly modified for this purpose, such as Crawford-Howell is. Although originally developed to compare neuropsychological score, I saw it to be used in papers for fMRI and VBM analysis more times (e.g. here). To run it on a smoothed/not smoothed fd, fc and fdc data (?) and then only run FWE correction using CFE on this-way-obtained uncorrected p-values?
I am openminded to any other ideas how to perform statistics in those specific cases, this is just what I came with.
Another degree of freedom in my case is using whole-brain fixels vs only fixels within some not-tract ROIs (means I have masks of areas subsequently resected, obviously not restricted to the specific tracts of WM). How does CFE correction works when I call fixelcfestats with -mask parameter? Is it limited on connectivity in this area or whole-brain fixel-fixel connectivity that I calculate before running stats? My point is (not specific only to Single-case control studies), does this connectivity information from outside the mask participates on correction?
Thanks for your help!