ICA removing blinks from EEG

Discussions
,
1

Hello!
Sometimes eye artifacts are regular, well visible and disappear from the record after the removal of one component of the ICA. But sometimes the removal of eye artifacts is difficult.

  1. How many ICA components do you think can be removed to achieve one type of artifact removal?
  2. If the eye artifacts on the recording are irregular and not very expressed, is it worth looking with much effort for a component, which somehow related to them, or is it better not to remove any component and decline epochs with artifacts on the next stage?
2

How many ICA components do you think can be removed to achieve one type of artifact removal?

We don't have enough expertise with ICA and EEG within the Brainstorm team to give a comprehensive answer to this question. There are too many parameters to give a number of components per artifacts: e.g. the number of electrodes, the types of ocular artifacts you are targeting (blinks are easier, other eye movements more complicated to understand and fix).

You could try asking the same question to experts in the EEGLAB community, or the authors of the EYE-EEG plugin for EEGLAB.

If the eye artifacts on the recording are irregular and not very expressed, is it worth looking with much effort for a component, which somehow related to them, or is it better not to remove any component and decline epochs with artifacts on the next stage?

There is always a trade-off between the amount of data you reject and and the amount of data you need for your analysis.
If you can reject all the trials with ocular artifacts and still have enough good trials left for obtaining decent results and significant statistical results in your analysis, then this is possibly a good option. This way, you'd be sure there is no bias introduced in your analysis from cleaning your data with ICA (as with ICA or SSP, it's always difficult to claim you are applying exactly the same type of correction for all the participants).
The main reason for which you would prefer fixing the artifacts over rejecting them is because of a lack of good data for the analysis.