Preprocessing EOG channels before ICA

[ah5005]

Hello,

I was having trouble removing blink artifacts with ICA, so I appended the EOG channel data to the EEG dataset. I demeaned without detrending (to keep the noise amplitude high), and accidentally did not re-reference them to mastoids like the EEG channels.

(As an additional detail, I am segmenting all the data, EOG and EEG, into trials and concatenating them, before running ICA.)

Miraculously this worked really well. I went back to re-reference the EOG data and the eyeblinks returned. What are your thoughts on including EOG data without re-referencing? Will this negatively impact ICA?

As a sub-question, if I don’t use these channels as EOG, I could merge them with EEG channels and attempt to recover brain data, though I’m not sure how noisy this will be – is this feasible?

Thank you so much.

[cbrnr (Clemens Brunner)]

See this entry in Makoto’s FAQ: Makoto's preprocessing pipeline - SCCN

It is recommended that your EOG channels share the same reference as all the other channels entering ICA. This doesn’t mean that ICA will not work if you use bipolar EOG channels, but in general this should not be necessary and avoided (because you decrease the data rank). If it didn’t work at all, it is likely that you did something wrong (maybe during the re-referencing step).

[ah5005]

Great, thanks!