Dear Aya,

the best approach depends strongly on what you are trying to do.

One way to avoid eye blink artefacts in your data, is simply throwing away all data with eye blinks. That mean, if you have a trial-based approach, you'd simply throw away all contaminated trials. The problem here is, that you'll likely lose a lot of data.

Another common approach is using ICA to clean easily identified artefacts. However, as you wrote "without modifying the signal", this is a problem. ICA will definitvely modify your data, albeit to the better.

A more advances approach that has recently been demonstrated is projecting your data into source space to obtain a cleaner signal (See Hipp & Siegel, 2013, Frontier in Human Neuroscience).

tl;dr: If you are fine with modifying your signal, use ICA. Otherwise throw away all contaminated data segments.

Good luck,

Julian

Thank you Julian,

Sorry I meant "without loosing segments" instead of "without modifying the signal".

In fact, I used ICA decomposition and throw the components that contain artefacts automatically using EEGLAB in MATLAB.. but these eye blink remain in the signal..

Thank you I will read the paper.

Dear Aya, 

Could you please load a topographic plot of this blink? 

I find it a bit weird that Fp1 and Fp2 are going in the opposite direction than the rest of the channels. Maybe this is not a blink, and this is why you fail to remove it? 

In any event, if this is a pattern that repeats throughout the data and you wish to remove it, it should be captured by the ICA components. You could browse the other components until you find the one that captures this activity, and then remove it.

Roni  

I have a limited experience in EEG signals ..

Yea this activity appear always in the same way. Fp1 and Fp2 in an opposite direction than the rest of channels..

If I run ICA in EEGLAB, how can I know which component contain these activities??

I'm not sure exactly how this is done with EEGLAB - I usually work with FieldTrip. But according to the waveform picture, you should see a component for which the topography goes positive (blue) in the two anterior channels and positive (red) in the posterior channels. In the waveform of the component you should see a peak in the same times you see this artifact.

You could try to remove components one by one, and see which ones removes this artifact.

You can also load a picture of the first few components, and I'll try to help.  

Ah okay, I found the component that removes the artifact.

Thank you a lot

In my opinion, ICA is not an optimum solution for cleaning EEG signal from an artifact, especially EOG, because is removing segment and consequently removing contaminated useful data. I prefer using EEG correcting or averaging techniques such as Gratton technique. 

By the way, I would like to ask, how many electrodes you have used and how long is you your recording?

I want to add one more thing about EEGLAB, you have to check the location file before processing your data, as some times the location file that provided by EEGLAB for many EEG electrodes location is not accurate.

Thank you all for you advices.

Hossam, I'm using 19 electrodes, and the record is about 10-11 minutes. Yes I uploaded my channel location file in EEGLAB.

Please could you give me some tutorials that explains this Gratton technique?

Thanks

Hi Aya:

ICA is a reasonable method though not perfect..Apart from removing the components related to Eyeblink I guess you also set the ICA weights(ie the unmixing matrix) corresponding to that component to zero before reconstructing to obtain the clean EEG channels

Some discussion to Gratton technique at EEGLAB list

http://sccn.ucsd.edu/pipermail/eeglablist/2012/005734.html

http://link.springer.com/article/10.3758%2FBF03209415

Thanks

Navin

You can find useful information on the EEGlab Wiki!

I can see from your image that the trigger codes are all different. Do you know what each individual trigger represents (i.e. 2200.1492 is a 'frequent' and 3010.0647 is a 'rare')? If so, the following link will help you in the process of select your events.

https://sccn.ucsd.edu/wiki/Chapter_07:_Selecting_Data_Epochs_and_Comparing

If you do not know what these triggers represent, try to deconstruct it from the stimulus presentation sequence which should have a log file.

Good luck

V

Hi ALL,

I do not have EOG data and am using 28. And when I apply the ADJUST on ICA-output data, I get an error: “ EEGLAB error in function_compute_GD() in line 68”. Any idea why do I get such an error?

However, when I go to the edit menu and specify channel locations for 28 electrodes, I receive ADJUST output. How could I validate if the ICs are correct or not??

As a newbie in this field, EEGLAB as a whole seems to be quite challenging for me!!. SO, if you could clear my doubts and questions, it would be really helpful for me.

Eagerly waiting for your reply.. 😊

Thanks !