Thresholding and different versions

Dear MRtrix team,

I would be grateful for your advice on 3 questions ?

  1. Does sift2 negate the need for graph thresholding in connectomics ?

  2. I have processed a dataset using the last 2 versions of mrtrix3 is this likely to cause problems and if so for what functions particularly ?

Thanks so much !


1 Like

Hi Peter,

  1. In short: yes (in my opinion). I wrote a little on this topic in the discussion section of this paper; there’s now some further thoughts in the discussion section of this recent paper.
    It’s also important to disambiguate between applying a threshold to low-density connections, and binarizing the connectome (as some do the former but not the latter). I’d argue against both: a ‘robust’ weighted network analysis should not experience a significant fluctuation in results if low-density connections are truncated to zero, so doing so is just introducing an arbitrary parameter.

  2. Depends on precisely which versions you used, and what functions you’re using. Although we have version ‘tags’ (e.g. we are now at 0.3.15), these actually only refer to specific points in time within a constant stream of updates on the master branch. So unless you manually ‘check out’ one of those tags, the precise point in time at which you update will influence what code you have. This will change once the software is properly released: many users will download / check out a specific version that remains stable for the duration of their analysis. Theses kinds of issues are precisely why I went to the effort to produce a changelog :grin:. I’m happy to give more information about any specific change if you’re concerned whether or not it necessitates re-running the analysis without mixing software versions.


1 Like