Connectivity tool

Discussions
1

Dear all,
I would like to have some advices on the use of the connectivity tool.

After obtaining the source maps I created a few (self-made) scouts according to my significant activations. The scouts were: rDLPFC, SMA and PPC.

Then, I run the “downsample to scouts” tool and run the bivariate granger causality NxN to get how the above scouts are causally linked.

The results are nice but I wondered this could be due just to the fact that I put only the scouts of interest according to my results.
I also tried to run it on the whole Desikan-Killiany atlas but results are quite hard to interpret.

My questions are:

1)Did I follow the right procedure?
2)Is there a minimum of scouts to use in order to get good results?Do I need to (randomly?) add some additional scouts?
3)What about the Maximum Granger Order model value?May I just leave the default value?
4)How should I describe the connectivity procedure in my paper? How can I report connectivity results and figures in my paper?Is there some recent paper that used the connectivity analysis on brainstorm ?

Thank you very much

Giovanni

2

Hi Giovanni,

  1. Yes, this sounds correct. Some comments:
  • Instead of using the “downsample to atlas” process, which is more designed for full cortical segmentations, you could use the process “Extract > Scouts time series”, and then use the signals as the input of the connectivity process.
  • In the Desikan-Killiany atlas, the regions are too large to be functionally homogeneous: averaging the source activity over large regions causes most of the interesting signal to be lost…

  1. All those methods are bivariance. All the pairs of signals are processed independently, the number of regions does not influence the results you get. We know that this is not the correct way to do this type of analysis, multivariate approaches are in development as well.

  2. All the functions related with the functional connectivity analysis in Brainstorm are still undocumented: incomplete theory, in development, not tested, subject to future changes.
    The Granger causality estimation particularly needs some more work to identify what is really significant in the results we obtain. Correlation and coherence are simpler metrics that can lead to less random results.
    If you still want to use this method, yes, use the current default value.

  3. For this: I don’t know… Coherence and correlation are working well, some people are probably using them now, but I have no reference to give you.

Cheers,
Francois