Permutation test on connectivity

1

I was trying to run test between connectivity files and this is what I got: “statistical tests on connectivity results are not supported yet”.
I was wondering: how long would it take for you to fix this? Should I wait a little bit or should I try to do the statistics somewhere else?
Thanks

2

I will let @Sylvain, @leahy and @hossein27en address this question.

3

In principle, the connectomes (ie the connectivity matrices) can be handled as the other statistical objects in Brainstorm (time series, time frequency and source maps, etc.). Preference would be to run permutation tests.
So @Francois and @MartinC, I suggest this blockade be removed.

4

I already wrote this code a while ago. It is possible to run permutation tests on 1xN and NxN connectivity matrices. The message “Statistical tests on connectivity results are not supported yet” is a only a warning, it doesn’t prevent the output file to be generated. We’ve been waiting for decisions to be made by @leahy and @Sylvain about a “default processing pipeline” for statistics on connectivity results.

Our tutorial pages are still empty on all these topics:
https://neuroimage.usc.edu/brainstorm/Tutorials/Connectivity
https://neuroimage.usc.edu/brainstorm/Tutorials/Workflows
https://neuroimage.usc.edu/brainstorm/Next#Connectivity

5

Hi everyone!
I was wondering the same question, more specific, the best statistic test for the Brainstorm’s Imaginary Coherence.

The iCoh that brainstorm computes is a kind of squared or normalizated measure(not the Nolte’s calculation), right ?
So I was wondering if it has a normal distribution?, in order to use parametric test or if is it better to use non-parametric test for this measure.

When I compute classical (squared) Coherence I use non-parametric test (like U-Mann) cause literature shows that coherence has not a normal distribution, unless you correct it with Fisher’s Z-transformation, in that case you can use parametric test.

But in the case of Brainstorm’s Imaginary Coherence I do not know what to use, any suggestions?

Thanks in advance

7

You can evaluate whether the iCoh measures are normally distributed by running the histogram routines of Brainstorm once you have computed the iCoh arrays (disclaimer: I haven’t tried if these latter work on this data type but in principle they should). You can also export the iCoh measures to Matlab and run a test for Gaussianity.

The safest approach, although more computationally intensive, is to resort to non parametric permutation tests.

8

I’m gonna try that!
Thanks!