Unfortunately ICA will not work in your situation because you only have data from one electrode (if I understand correctly).  ICA can only create components equal to one less than the total number of electrodes that you have recordings from.  Thus only one component in this case.  Typically you would need at least 16 electrodes to get a relatively clean component for eye brinks.  More if you are also trying to correct for horizontal eye movements.  I'm not sure what you are doing with the EEG from Fz but you might be able to see eye blinks manually in Fz.  If so you could try to manually eliminate them if you are trying to make an ERP.  If you are doing a frequency analysis you might not have to worry about ocular artifacts depending on the frequency band of interest.  But to be honest there is not much you can do without multiple electrodes or at least one electrode underneath the eye.

no in order to correct for ocular artifacts you must have additional electrodes placed in the vicinity of the eyes.

Thank you for your quick reply. I understand that it is possible if we have data from at least two electrode. However, is it possible to detect automatically ocular artefact in a fragment of a signal?

In general, if you can see the artifact with your eyes, you should be able to detect it. Check ICA (independent component analysis). You can use ICA to both detect and remove the artifact from your signal. 

Unfortunately ICA will not work in your situation because you only have data from one electrode (if I understand correctly).  ICA can only create components equal to one less than the total number of electrodes that you have recordings from.  Thus only one component in this case.  Typically you would need at least 16 electrodes to get a relatively clean component for eye brinks.  More if you are also trying to correct for horizontal eye movements.  I'm not sure what you are doing with the EEG from Fz but you might be able to see eye blinks manually in Fz.  If so you could try to manually eliminate them if you are trying to make an ERP.  If you are doing a frequency analysis you might not have to worry about ocular artifacts depending on the frequency band of interest.  But to be honest there is not much you can do without multiple electrodes or at least one electrode underneath the eye.

I also agree with Robert's answer. With a single electrode you can only watch to raw data and remove manually the segments affected by this artifact. Another possibility that came now to my mind is to create a template of an artificial signal artifact-shaped. Using it, you could design a function in some Matlab-like language that compares the signal  that you recorded (previously pieced in short intervals) with your template. Likewise, you may use some similarity function you could automatically detect and remove the artifact-like parts of your signal. However, if your record is not very long, I would suggest you to do it manually.

You can delete artifacts using a single electrode. An example of the removal of the most common eye blinking artifacts can be found  in the attached file

Article Zastosowanie klasteryzacji do automatycznej detekcji artefak...

Marcin, given your paper isn't in English could you give a brief description of how your method works?  Does it generalize across multiple subjects? What did you do with the EEG once it was cleaned and how did you verify that it did not change the EEG signal relative to accepted practices such as eye blink rejection based on eye monitoring?  Thanks

Marcin, accurancy 98% - you mean sensitivity or specifity?

The signal produced by eye blink is in mV while EEG signal is in microV, i think is possible to remove dirty data with a simple threshold methods.

Another way may be the same method used to remove ECG from other signal. For example use a model trained on a recorded data, like ARX or Output Error.

No...we need other Electrode..

Change reference

I agree with Giacomo, if you have trials and a relatively controlled experiment (fixation on fixation cross - no continuous eye movement) you might be able to identify eye blinks or movements based on their high voltage - and subsequently reject trials that show a voltage greater than that of normal EEG (which should be EEG system, reference, subject specific, I guess). Anyway, if you identify a threshold based on observation of EEG on a small normal (i.e. no eye movement) EEG data segment, you can use it in an automatic fashion for this subject subsequently. However, if you measure EEG in a less controlled environment than you will probably have lots of eye movements and hence loose lots of data that way..

Hi Wojciech Sałabun.

Maybe you could try to use the AAS http://web.mit.edu/swg/ImagingPubs/Multimodal%20Imaging/Allen.NeuroImage.2000.pdf (Allen et al., 2000). Basically If you have an artefact that happens several times during your acquisition and you can consider it similar through the time, you can create a template using several repetitions and subtract it from your channel. We use this approach for EEG-fMRI, where we have the gradient artefact (caused on EEG by the fMRI acquisition). In your case it would be more similar to the BCG correction, where we have an artefact inserted on EEG due to the vascular pulsation related to the heart beats. In this case you would need to mark when you have the R peaks (ECG), which would be similar to the case you have the ocular-related artefact. So, If you can identify and mark these ocular artefacts, maybe you can use AAS and try to correct your data.

I hope I was clear.

Cheers

In the frontal electrode site the ocular artifact is very clear to find by visual inspection. you can remove it manually why do you prefer to use computation method to remove?

To ensure that you are not removing actual signal thresholding is typically done with a bipolar electrode subtraction between an electrode under the eye and an electrode over the eye like Fp1.   It is impossible to ensure that you are not removing signal without using electrodes above and below the eye.  While ERPs are in microvolts eeg signal (and also EMG) itself frequently moves in millivolts.  Likewise blinks can frequently occur coincident with ERPs and thus without a precise detection scheme you wouldn't know this.  Please refer to the chapter on artifact detection and correction in Luck's ERP methods book:  http://www.amazon.com/Introduction-Event-Related-Potential-Technique-Neuroscience/dp/0262621967 for additional details.  

Blink artifacts can be detected by a simple thresholding procedure, because eye blinks generate huge potentials. For example, a threshold of about 100 uV (change relative to baseline) might work.

Horizontal eye movements will be a problem. You would need a bipolar outer canthus montage for horizontal eye movements (= analogous to the montage mentioned by Robert, but here one electrode is placed left to the left eye and the other, right to the right eye)

Robert already mentioned that spatial filtering (ICA) is not an option you have, and I think that filtering in the frequency domain won't help either. You can remove some noise with a low-pass filter, but that won't get rid of eye artifacts. Notably, high-pass filters can distort potentials generated by eye blinks, so that it will become difficult to find them with thresholding procedures.

A more general question is what you want to examine. 

If I had a single electrode, I would not place it at Fp1.

Too many artifacts and too little brain activity at this site (= poor signal-to-noise ratio)

cheers

Tobi

Thank you all for extremely precious advices. Unfortunately, my EEG device is designed only for Fp1 area. Now, I am interested in improvement this signal (e.g. identify EEG artefacts, remove, repair etc.) New comments are welcome.

I have developed an open source matlab toolbox for the automatic EEG artifact rejection, as published in: http://www.mdpi.com/1099-4300/16/12/6553

If you are interested, please visit: http://nadiamam.wix.com/eawica

This is in response to what Wojciech Sałabun wrote.

We had the same problem.  Our system was designed for FP1, FP2.  We made external connections to make it possible to capture signals from C3, C4.   We connected a wire from C3 to the device.  Tapped ground at A2(right ear lobe).  It is possible to rewire it and use it for tapping other signals. 

We  re trying to get single channel data to detect arm movement. Has anyone done discrimination between the left and right arm movement with single channel?