Dear MRtrix experts,
I’m trying to compute differences between 2 groups in structural connectomes.
According to the documentation provided in the 1st Mrtrix workshop, once individual connectomes were generated, some small modifications of these individual connectomes were suggested in preparation for statistical inference.
Some of the steps suggested included calculating the group mean connectome (after scaling by mu), sorting the edge densities (excluding self-connections) and choosing the 80th-percentile value.
The question is, if I have 2 groups, should I calculate the mean (and sorting the densities etc ) for each group separately?
I was also wondering whether this procedure is up to date and whether new documentation is available on how to proceed in group connectomics.
I never got around to writing a script for automating these processes, primarily because we have become increasingly pessimistic about the application of the TFNBS / NBSE method, to the extent that it was not even presented in the most recent workshop. We will hopefully have some published work in the future outlining the issues that we have encountered.
Nevertheless, if you do want to pursue this avenue:
The question is, if I have 2 groups, should I calculate the mean (and sorting the densities etc ) for each group separately?
Absolutely not. This could result in having edges that are zero-filled in one group and contain valid data in the other group. Such edges would receive enormous t-statistics. The purpose of that step is to indicate which edges should participate in the statistical analysis and which should not. A -mask option in connectomestats would actually be more sensible in that respect - note this would still require deriving such a mask - but this shows how that mask couldn’t possibly be different between the two groups involved in the analysis.