Boundaries in continuous EEG and PSD calculation


I import EEG data analyzed in EEGLab in Barainstorm to do PSD (Welch) analysis. Some parts of the continuous EEG was rejected and the boundaries visible in Barainstorm.

Is it OK to include these parts in PSD calculation? Are not they automatically excluded from PSD calculation?


If you want to exclude these bad segments from the PSD computation, mark the event category as bad:
In the Record tab, menu Events > Mark group as bad.
Alternatively, you can rename the events and add the tag "bad_" in their name, it would have the same effect.

Thank you Francois.

Hi there,
I've come across a similar issue, maybe my first question should be 'Does boundaries affect PSD and further analyses'?

I believe the boundaries are causing issues. On the raw data, the boundaries appear multiple times in a single second, and they are 'stacked' as seen in the record window - the boundaries constantly overlap. After plotting raw data PSD, the power reaches x10^4 figures, which I believe is caused by the boundaries (correct me if I'm wrong).

I tried using the method above by labelling boundaries as "bad" but that leaves no actual data to use and the PSD fails. I tried introducing my own "BAD" event group and manually selecting portion of time where just the boundary line is seen in the data - PSD also failed with the same error message. I've tried the "Detect other artefacts" option, ended up with the same result.
I was wondering if anyone had any more suggestions for how to remove boundaries or is this data unusable? (This is scalp eeg data, intention is to do source localisation, etc).

Thank you!

If these "boundaries" are separation between different epochs, you shouldn't be processing this file as a continuous file. You can't compute a PSD or apply frequency filters on contatenated epochs.
Use the menu "Import MEG/EEG" instead of "Review raw file", and if you don't have the epochs information available in the dialog window, use the option "Split" to import the input file as separate epochs.

To compute the PSD, it will be complicated to have multiple overlapping estimator window in each epoch.
Select all the epochs in Process1 and run either the PSD process with the time window = the epoch duration, or the FFT process with the option "Average". Both should give the same results if the options are configured correctly.

Hi Francois,
Thank you for your message. I admit the process is more complex than I thought.

I have taken your advice and re-imported my epoched data using "Import MEG/EEG". During import I selected the "split" option and then, "boundary" event, which split the raw data into 352 files. Once imported, raw files of about 4-5 seconds each contain about 10 boundary events, each. But boundary events this time are single events named "65281".

I had a look on Brainstorm under Import Epochs:
I could not find any further advise on removing the remaining boundary events.

I can't think of any other way to remove 2 separate sets of epochs. I believe I will only be able to use 1 second epochs from this data.

Thank you for your help!

What I had in mind was to import with the split option = the duration of one epoch.
So that in your database you have one file = one epoch, and no problems of discontinuities in your signals.

Alternatively: Use the menu "Review raw file" to create a link to the EEGLAB .set file. Then Right-click on the link > Import to database > Select event "65281" with an epoch time: [0ms - duration of each epoch].

1 Like