Bad looking ERP pattern

Discussions
1

Hi all,

First, thanks for this amazing program and really clear tutorials that were helpful for a first-time analysis of EEG data.

I am trying to run ERP comparison on the data I collected and I encounter a problem at the grand average step that I don’t understand so I was wondering if someone ever encountered the same problem and can provide me some help with it. My issue is that for some electrode in some experimental conditions the signal looks weird (see the purple line in F1 and F5 electrodes and the green one in the Fz electrode on the figure attached, which correspond to grand average for each of the three experimental condition of my experiment).

bad%20looking%20ERP%20pattern
bad%20looking%20ERP%20pattern1919×1015 373 KB

The experiment was only one run per subject and the three experimental conditions were randomized so I assume that this problem is not due to one particular bad electrode otherwise this weird pattern would have been the same whatever the experimental condition and won’t have been encountered in different electrodes.

Here are the steps of my analysis :

  • Filtering data (0.5 – 70Hz + Notch at 50Hz)
  • Detecting and rejecting bad electrodes using Power Spectrum Density and manual inspection of data, then interpolating bad electrodes with the default parameters
  • Detecting ocular artefact with an ICA and rejecting the components corresponding to blinks and saccades
  • Segmentation of trials with DC offset correction using almost the entire of the fixation cross presentation (-1500 ms; -2 ms) as a baseline
  • Manual rejection of bad segments
  • Baseline normalization using ERD/ERS method with the intervall between -750 and -250 ms of the fixation cross (i.e., baseline) to avoid edge effects
  • Weighted average for each subject (because of different amount of trials in each experimental condition)
  • Average folders across subject (grand average)

May I have done something wrong ? Do you have any idea of why I have this kind of ERP pattern ?
Thanks a lot for your help!
Lison

2

Hello,

Your analysis steps look good except for the ERD/ERS normalization. This is not something that you should be using on EEG recordings before averaging, it is meant to normalize time-frequency maps.
What edge effects are you referring to?

The baseline correction you typically want to use on your recordings is the one that is embedded in the data import (option “Remove DC offset”), that you could also achieve with a high-pass filter: http://neuroimage.usc.edu/brainstorm/Tutorials/Epoching#Import_in_database

Francois

3

Thank you François for your really quick anwer.

I also wanted to run a time-frequency analysis so I followed the tutorial describing it and mentionning the high edge effect that have to be taken into account fr the baseline normalization (http://neuroimage.usc.edu/brainstorm/Tutorials/TimeFrequency), so that is what I did, and I assume that is were I get confused between ERP and time-frequency analyses procedures.

So, if I understood well now, the ERD/ERS normalization have to be done:

  • after and not before the averaging,
  • only for the time-frequency analyses and not for the ERP ones
    Right?

Thanks again,

Lison

4

Correct.
Go back to the tutorials in case of doubts.
You have a summary of recommended analysis pipelines on this page: http://neuroimage.usc.edu/brainstorm/Tutorials/Workflows