1

I have 29-channel EEG data from overnight sleep recordings (~8 hours of EEG data). The EEG data is not continuous as I have paused it if the subject needed a loo break (~5-10 minutes) or to check the impedance (~2 minutes). This inserts a ‘boundary event’ in my EEG data. On some occasions, I have also reapplied a couple of electrodes if the impedance was high. In addition, sometimes I have had to stop the EEG recording and start again due to some technical reasons, because of which I have two EEG files for the same subject during the same night i.e. same session.

In my experiment, I have 15-20 events overnight and plan to use 1-2 minute data before each event for preprocessing and analyses. I would like to use ICA to correct for artifacts but before I do that, I would like to know if I need to split the data at each ‘boundary event’ and process them as separate EEG files? Or can I just consider the EEG recording from each subject as a single EEG recording and perform ICA (I can append the EEG files in cases where I have two EEG files per subject in the same session)? Any suggestions will be highly appreciated as I am very new to EEG and MATLAB.

EEGLAB User
  • 11
  • 2

1 Answers1

2

You can find a detailed comment on your issue here https://sccn.ucsd.edu/pipermail/eeglablist/2014/007380.html

Changing the impedance of the electrode introduce non-stationarity of the signal, that could affect ICA. However, ICA seems robust to this kind of violation, so you could concatenate all the sessions and run the ICA.

Rather, an issue could be if the electrodes moved during the breaks (ICA is a spatial filter, so spatial releationship are very important. see also here https://sccn.ucsd.edu/pipermail/eeglablist/2016/011878.html).

As a personal comment, I would use an "empirical" approach, to get a better grasp on the data and on the problem. In EEGLAB you could easily write a script and perform two parallel analysis, keeping all equal but the ICA decomposition (one on the concatenated data, one on the data separated for sessions). Then you can compare the difference on the data at the last step. In this way you can appreciate how the different strategy affected your results, and if this is really relevant.

Giorgio
  • 41
  • 1
  • 5