I am phd student in cognitive science (cognitive modeling).
I have some questions:
My research is about finding some features on resting state EEG of two groups .
2)I have 200 Seconds resting state EEG for each case.after pre-processing I want to choose 5 segment(block) of EEG data with 10 second duration for each.
3)then I want to measure ROI(Scouts) activity by source localization using brainstorm.
then I want to find effective connectivity between ROI(sources) for each segment.
what do you suggest ? GC or transfer entropy or something else ? can I do that in brainstorm?
I tested PTE & GC N*N in brainstorm
5)after that I want to compare all case/segments in two groups
which statistic method do you suggest ? Can i do that in brainstorm?
6)Do you suggest another method? Is there any research paper that used brainstorm for measuring connectivity ?
Hi!
I would use Amplitude Envelope Correlation: it has a correction for volume conduction (the ortogonalize signals option) and it has probed to be robust and better than many functional connectivity measures.
for statistical analysis I suggest you the two groups permutation test,
sure you can do in BS everything you wrote in last post
Yes, in that case you can use Granger's Causality, but I think it dependes what you want to know, there is another measures that give the information flow direction like Imaginary Coherence, but I'm not sure about how useful or valid is to get the causality in resting state...
As another member said, Amplitude Envelope Correlation is a pretty good method to compute the Resting-State connectivity when you have recordings that are not short (like 500ms). For your case, it looks great.
To compute the Effective connectivity there are other approaches. Imaginary and Lagged coherence are suitable algorithms plus Granger Causality. All of those measures are already in Brainstorm.
Granger usually works well when you don't have too many channels.
In case you want to construct sources, please don't compute them by absolute values of dipoles since it damages the frequency characterization of signals. A simple average should work well.
just curious, but when you say "recordings that are not short (like 500 ms)", are you saying that 500 ms is okay for AEC, or that it's considered to be too short for AEC?
Dear hosein
Thank you for your response
You told me that I should use average instead of absolute values when I want to compute sources. what does it mean? Where is this option in brainstorm?
And I want your idea about phase transfer entropy?I decide to use this measure for connectivty
I want to write my pipeline here and I want to know it is correct or not:
Import raw EEG data for each subject
Divide each raw data to 15 blocks : each of them 10 second
Use some preprocessing like filtering , removing artifacts and ...
Choose 10 blocks for each subject
Compute headmodel using openMEEG BEM ( I used computed headmodel of first subject to for all subject - due to using default anaotmy)
Use no noise modelling
Compute data covariance for each subject
Compute sources : Minimum norm imaging , sLoreta , constrained for all of selected blocks
Define ROIs ( Using desikan killiany atlas)
Measure connectivity PTE for sources of each block (between ROIs)
10)statistic ( using permutation test between two conditions - Process 2 - copy responder's blocks to file A and nonresponse's blocks to file B)
Divide each raw data to 15 blocks : each of them 10 second
Do not split your recordings before the preprocessing. The filtering should always be done on the continuous files, as illustrated in all the online tutorials. If you import shorter blocks of 10s, do it from the cleaned continuous files.
Compute data covariance for each subject
You don't need a data covariance, this is only needed for the LCMV beamformer.
Compute sources : Minimum norm imaging , sLoreta , constrained for all of selected blocks
I want to know when you told me thay I should use non normalized source maps , you suggested to use current density map instead of sLoreta or something else?
It looks like it. But there are lots of possible options, I can't guarantee that everything you do is correct. Make sure you follow the guidelines we provide. This page contains a summary a various recommended pipelines: https://neuroimage.usc.edu/brainstorm/Tutorials/Workflows
current density map instead of sLoreta or something else?
Try to use the current density maps instead of sLORETA.
See the recommendations in the page I linked.
The connectivity graphs are rendered with JOGL, not with the Matlab OpenGL renderer.
This prevents us from making screen captures using the the Matlab functions.
You can use the tools from your operating system for making a screen capture. On Windows or Linux: press the key PrintScreen (together with the ALT key for capturing only the current figure), then paste it (CTRL+V) in any image editing software.