Dear Community,
In my experiment, I have separetely registered EEGs for 6 conditions with two registrations for each condition, which gives 12 trials per subject.
For each subject, all the trials are recorded one after the other, in the same experimental condition. Thus, I would like to perform the preprocessing (filtering, bad segments, bad channels, ICA) for each subject on the 12 trials data at the same time by concatenating the trials.
I have read how to do so on other posts. However, my question is about the "de-concatenation" of the cleaned concatenated file, into the 12 cleaned trials. My goal is to average (by subject) data from the same conditions together, then perform the grand average over subjects per condition, and also perform source localization.
Is it possible to deconcatenate the cleaned data into the same time intervals as the original trials?
If so, what would happen to the ICA mixing matrix that is kept in the channel file after ICA, and needed for the forward model?
Thank you very much for your time and help.
François
I'm not sure I understand what you mean with "two registrations for each condition".
For each subject, all the trials are recorded one after the other, in the same experimental condition. Thus, I would like to perform the preprocessing (filtering, bad segments, bad channels, ICA) for each subject on the 12 trials data at the same time by concatenating the trials.
Filtering and (ICA analysis preferably) should be done on continuous recordings, not on epoched files.
Filtering short epochs may cause important artifacts at the beginning and the end of the epochs. Concatenating artificially discontinuous data blocks does not make this problem disappear: the discontinuities will create similar artifacts, around the junction between two epochs you can't reliably use the filtered data.
If you really want to do the ICA on concatenated epochs, and found justification in the literature that you can do this safely, you don't have to explicitly concatenate the files first: in Process2, select all the epochs in FilesA AND in FilesB. It will concatenate the files you selected in FilesA, compute an ICA mixing matrix, and apply in to all the files in FilesB.
Bad segments / bad channels: this doesn't have much to do on concatenated epochs.
My recommendation is overall: do not concatenate your files in the first place.
If you still want to do it, then you need to use event markers to re-epoch your data with the import options window, as illustrated in the tutorials.
Hello François,
thank you for your answer.
I made a mistake in the use of the word "trial", I meant "run" or "recording". Thus I guess my explanation was a bit confusing, please excuse me. I do not want to concatenate epochs but raw files.
To make things simpler, let's say I have 12 different EEG recordings of the same subject in the same experimental conditions. I want to concatenate the raw files of those 12 EEG recordings, to perform the preprocessing (bad channels, ICA) at once on the whole concatenated file.
However, after the preprocessing, I would like to cut down the preprocessed concatenated file into the 12 different (now cleaned) recordings.
Is this possible? If so, what would happen to the ICA mixing matrix that is kept in the channel file, and needed for the forward model?
I hope I made myself clearer.
Thank you.
François
Ok, thanks for the clarification. Then I think you can proceed as I indicated:
you don't have to explicitly concatenate the files first: in Process2, select all the epochs in FilesA AND in FilesB. It will concatenate the files you selected in FilesA, compute an ICA mixing matrix, and apply in to all the files in FilesB.
The same ICA mixing matrix would be copied to all the channel files for the recordings in FilesB.
For bad channel identification: it's better to proceed separately for each file, in case an electrode goes bad when the subject moves between two acquisition runs.
For bad segment identification: there is no advantage of proceeding all the segments at once.
1 Like
Thank you François, I tried it and it worked well.
Thanks also for the additional advices on channel identification and bad segments!