Best practice for EOG artifact detection in MNE-BIDS-Pipeline without dedicated EOG channels

I’m using the MNE-BIDS-Pipeline on EEG data that doesn’t have dedicated EOG channels. I want to automatically select ICs corresponding to EOG artifacts during the ICA step. We alo conduct studies with children, where horizontal eye movements are more problematic.

I’m considering using frontal EEG channels to simulate EOG channels:

  • Fp1/Fp2 for vertical EOG (VEOG)
  • F9/F10 for horizontal EOG (HEOG)

In the MNE-BIDS-Pipeline config, I’ve thought about these options:

  1. Using all channels:
    eog_channels = ["Fp1", "Fp2", "F9", "F10"]
  2. Using a subset of channels:
    eog_channels = ["Fp1", "F9"]
  3. Creating a bipolar HEOG channel and using it with Fp1/Fp2:
    eeg_add_bipolar_channels = {'HEOG': ('F9', 'F10')}
    eog_channels = ["Fp1", "Fp2", "HEOG"]

What would be the best practice for simulating EOG channels and configuring the pipeline for optimal EOG artifact detection, especially considering our work with children’s data?

Thank you for your insights!

If you don’t have EOG channels in your data, but EEG channels in a suitable setup (like F9, F10 for an HEOG), doing a bipolar referencing of these channels is a good idea.

Using your Fp1 channel as a VEOG may work, but since you cannot re-reference to an electrode right under the eye, this may be a bit more difficult.

You could also try to visually identify what vertical eyeblink ICA topomaps look like in one participant of your dataset, and then apply a method like corrmap to the rest of your data. Or you try to use ICAlabel.