Hi. it would be most useful (and even necessary) to look at the time courses (waveforms) of the ICs. None of the spatial patterns that you show look precisely like blink patterns that should be symmetric and of the same polarity on the left and right frontal scalp. We have discussed the blink artifacts at some length in our MEG–EEG Primer (by Hari & Puce), Oxford Univ Press 2017; maybe it would be helpful to have a look at our pictures.

As Henri mentioned, definitely remove other, big, obvious artifacts before running ICA. Just by looking at the topography of IC2 I'd guess it's a saccade. You don't need the subject to look to sides a lot to get this component in the ICA, because ocular artifacts are very strong compared to the brain activity. I wouldn't remove IC3 (only based on your plots). IC11 may potentially be a noisy frontal electrode - look at the time courses. IC1 needs to be inspected in time courses as well. You can also remove the ones you're sure about, see the "updated" signal and try again with and without electrodes you're not sure about. See how the whole signal changes when you accept/reject a component. In the end it's a decision-making process. Good luck!

As Henri said, try to clean your data before running the ICA. It will benefit from less noisy signals. IC2 might be a saccade, but I would first re-run ICA after obvious artifacts removal (huge head movements, repeated blinks, yawns, etc.). Secondly, I would take a look at this tutorial https://labeling.ucsd.edu, and, finally, exclude only ICs with clearly recognisable artefacts: you don't want to remove truly brain activity with this process.

As Riitta Hari says, look at the continuous ICA activity (in EEGLAB: plot component activity). My guess from the spectrum and topo is that IC2 is a particular heart-related artifact that seemingly happens more often if the elastic cap is too tight. You can check it by looking at the component activity or the EEG activity accounted for by the component (in EEGLAB, remove component, then don't save anything but just plot). If it's a pretty clear bump occurring between 1 and 1.5 x per second (i.e. 60-90 BPM), it may well be related to the heart. The time related data in the plot doesn't show this though.

Following on from Luca Bischetti , rather than just using the ICLabel website for training, you can use the ICLabel EEGLAB plug-in, which uses the data collated from the ICLabel website and machine learning to label your ICs.

It gives a % likelihood of classification and also rather helpfully plots the topography alongside the time series, continuous frequency and spectrogram data.

EDIT ~ As this has been recommended a few times now, I wanted to also say that you can have the ICLabel plugin automatically remove components based on percentage alignment likelihoods, or you can also still do it manually by component number.

I think it's hard to suggest you remove any component by inspecting the images you provided and except the last component (IC11) which seems to be frontal electrode artifact. You need to inspect the time course of the IC and follow the proper preprocessing pipeline before applying ICA. What I found that no automatic artifact rejection algorithm in eeglab can detect all sort of artifacts though their performance is quite good it's better to eliminate artifactual component by visual inspection as well

Ty Lees Riitta Hari Henri Pöntynen Magdalena Matyjek Luca Bischetti Michiel M Spapé Muhammad Adeel Khan

thank you so much to all of you for your answers and for your time. your answers help me a lot and they were useful. sorry that I couldn't thank you all sooner.

As a suggestion, I have been getting quite good results with the seven-step procedure describe in https://sccn.ucsd.edu/wiki/Chapter_01:_Rejecting_Artifacts. (Rejection based on independent data components) in which two ICAs are run. As noted above, examining the time course of the ICs is crucial. For example, from my experience, IC2 seems due to typical horizontal eye movements. If IC2's time course shows the typical "boxcar-shaped" deflections due to saccades, then such is very likely the case. IC11 could be due to, e.g., sudden amplitude fluctuations at F7 resulting from impedance fluctuations or electronic-device interference. In cases like this, rejecting epochs (rather than ICs) based on a first ICA (not aimed at IC rejection) may be worth considering.

Hi, I recommend you to use some automated method for artifact recognition among IC components.

You can check my paper here: Article Online EEG artifact removal for BCI applications by adaptive...

There, the procedure for dividing components in artefactual and brain signals is well explicated. In case of doubts, please do not hesitate to contact me :)

If you don't know what the components are, don't reject them!