perform statistical testing for each fixel independently?

Dear MRtrix experts,

I am doing fixel-based analysis. I understand that the usual way is to use fixelcfestats to perform the statistical analysis, but in my special case, I hope to perform statistical testing for each fixel independently, without neighboring information or CFE (connectivity-based fixel enhancement), and get the statistics (like p-values etc). Could I do this with MRtrix? If so, how shall I do it?

Thank you very much in advance!


1 Like

Welcome Chenying!

There’s a couple of ways in which this could be done, requiring varying amounts of effort in implementing missing features:

  1. Use the vectorstats command, which is intended for unstructured data where statistical enhancement does not take place.
    You could take your fixel data and convert it to matrix data in text form using the mrdump command. The results of statistical inference would however also be in this numerical matrix form. If you needed to get the data back into the fixel directory format, I think the way to do that currently would be to exploit the Matlab functions provided for reading & writing the MRtrix image formats (i.e. load an existing fixel data file, overwrite the data with the output of vectorstats, save to a new fixel data file). I’d like for there to be other ways to do such a conversion, but they’d require development effort.

  2. Synthesize fixel-fixel connectivity matrix data where each fixel is simply connected to itself but nothing else. This requires an understanding of the fixel-fixel connectivity matrix format, which is not yet documented, but it could be done. CFE would still “be used” by fixelcfestats, it would just produce “enhanced” statistics that are identical to the (Z-transformed) statistics from the GLM.

  3. Modify the fixelcfestats command to be capable of selecting a no-operation statistical enhancement class. The code is designed to facilitate such a thing, and indeed connectomestats already provides such.