Hi all,
I just wanted to post my processing pipeline to make sure I’m not leaving any glaring errors. I’ll be using the default anatomy across all subjects.
1.) importing eeg for each subject/condition
2.) calculate shared kernel of sources for each subject(sLORETTA)
3.) average source maps from individual trials together within subject
4.) z-score normalize
5.) run connectivity(icoh) for nxn scout matrix
7.) average connectivity results across subjects
8.) split connectivity results into freq bands
Does this pipeline seem sound? Is the z-normalizing necessary? If it is not, I was only averaging together the trials to make sure the z-scores were comparable between subjects(each subject has a differing number of trials), so if I’m not using z-scores should I average trials together after or before computing connectivity?
Also if I wanted to calculate connectivity between these sources and emg channels, should I also average the emg trials together and perform a z-score normalization on the emg data before checking connectivity between emg and eeg source data? So both emg and eeg sources would be z-normalized.
Does this all seem normal?
Best,
Mike
3.) average source maps from individual trials together within subject
Within a subject, the sources of the average are the same as the average of the sources.
Use the sLORETA link for the subject-level average instead of re-averaging the individual trials.
https://neuroimage.usc.edu/brainstorm/Tutorials/Workflows#Average:_Single_subject
(For EEG, if you have only one acquisition session, you can consider you would have only subject-level average per experimental condition)
4.) z-score normalize
5.) run connectivity(icoh) for nxn scout matrix
If you normalize again the source maps, it would be like if you had not applied the sLORETA normalization, it's redundant.
https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Standardization_of_source_maps
It should not affect much the results of the connectivity analysis anyway. If this coherence measure is your only measure of interest, you could probably even use directly the non-normalized current density maps.
5.) run connectivity(icoh) for nxn scout matrix
You should not compute any connectivity measure on averaged results, only on single trials.
Also if I wanted to calculate connectivity between these sources and emg channels, should I also average the emg trials together and perform a z-score normalization on the emg data before checking connectivity between emg and eeg source data?
In Process2, put the recordings of all the single trials in FilesA, and the corresponding sources in FilesB (select the same files and just click on the button "process sources" on the right). Compute the measure AxB (A=emg signal, B=sources).
I'm not sure the amplitude normalization would have much impact on the coherence measure, but maybe I'm wrong, check it.
@Francois So correct procedure for connectivity would be to compute sources for each trial, compute connectivity for each trial, then average connectivity for the trials together for each subject? This value could then be compared between subjects condition?
And for the emg part, when you say put all recordings in filea you mean just the emg data right? And then all source recordings in fileb? Wouldn’t I then just run connectivity between these two? Why multiply them? To clarify, source data is only coming from eeg data. Emg data is from leg muscles and this data was not used in source reconstruction
Thanks for the help!
Best,
Mike
to compute sources for each trial
More precisely: Compute a shared inverse kernel for all the trials in the same folder, and use the "source links" as sources for each trial. Do not recompute the same source model for each trial.
https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Computing_sources_for_single_trials
then average connectivity for the trials together for each subject?
I can't say if this is better than estimating the coherence from the concatenated trials.
I've just pushed a modification of the Coherence processes to allow the two options more easily (update Brainstorm to get it):
Process: Add option to average connectivity measures (cohere, granger) · brainstorm-tools/brainstorm3@559834a · GitHub
And for the emg part, when you say put all recordings in filea you mean just the emg data right? And then all source recordings in fileb?
You EMG and EEG data are not saved in the same files?
If not, then indeed: FileA=EMG files.
The selection of the channels you want to use for the computation is done in the options of the Coherence [AxB] process.