I am getting conflicting advise regarding how to clean my EEG data:

1) Manually remove artefacts first and then apply digital filters


2) Apply digital filters first and then manually remove artefacts

The reason given for 1) is because artefacts are more visible and avoids accidentally accepting artefacts as EEG data.

The reason given for 2) is because it avoids accidentally rejecting EEG data.

Both reasoning makes sense to me and I am a little confused as to what I should do: 1) or 2). Anyone can advise? Thank you very much.

  • $\begingroup$ It depends on the nature of the artefacts. E.g. eye movements, jaw movements, clenching etc , produce so strong movements that entire trials including them can be excluded. So if you address these kind of artifices procedure 1 is the one to go. On the other hand, artefacts due to other neurophysiological sources like heart beat can be eliminated by other approaches (like ICA or beamforming), here you would apply filtering in advance and then try to remove these artefacts. $\endgroup$ – Irreducible Sep 20 '18 at 5:29
  • $\begingroup$ Yes, I'm starting to see it depends on the data type as well. I'm just starting out in EEG signal processing and starting to see the nuances that can make it complicated. Thank you for your comment, it was insightful. $\endgroup$ – Krithika Sep 20 '18 at 8:12

When it comes to preprocessing of EEG data, the first step is to filter the signal. Filtering is done such that it preserves information across different bands of the EEG data (alpha, beta, gamma, delta). Typically, many researchers use 0.5 - 50 Hz band in the first step (removes DC components). Note that the filter should be zero-phase such that no delay or phase changes are introduced in the EEG data on using the filter. One simple way is to use filtfilt command in MATLAB.

In addition, a notch filter is also designed to remove 60 Hz. You might be wondering why we need a notch filter at 60 Hz if we are filtering the original signal using a band-pass filter of 0.5 - 50 Hz. To answer this, we need to first understand the structure of the band-pass filter. If the band-pass filter is of lower order, then the transition bands are not very steep (or sharp). This in turn introduces some part of the 60 Hz interference into the band-pass filtered signal (which typically dominates in EEG data). So, to suppress the strong 60 Hz interference, a notch filter is also used.

After filtering the EEG data using a band-pass and a notch filter, you might want to address different types of artifacts that show up in the EEG data. Some of the them include EKG artifact, movement artifact, sweat artifact, etc. Each artifact removal requires a sophisticated approach such that the EEG data is preserved but the artifact is removed completely.

  • $\begingroup$ I didn't know that about the notch filter. I am using a band-pass of 2Hz-35Hz but I haven't used a notch filter for mains noise because I thought the band-pass would have filtered it out. I was also advised to look at the filtered data and raw data together to see if the artefacts in the raw data infiltrate the filtered data, and then decide how I should proceed from there. This is the likely approach I will be going with, so a combination of 1) and 2). Thank you for the answer, it was detailed and very helpful. $\endgroup$ – Krithika Sep 20 '18 at 8:14
  • $\begingroup$ Based on my personal experience, filtering alone will not be sufficient to remove artifacts. For example, the sweat artifact appears at 1-2 Hz frequency which falls within your band-pass filter region. $\endgroup$ – Maxtron Sep 20 '18 at 15:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.