EOG Denoising

Concerning processing components: filters, file load/save, visualizations, communication ...
Post Reply
amitpbarde
Posts: 7
Joined: Wed Sep 12, 2018 4:56 am

EOG Denoising

Post by amitpbarde »

Hello,
I’m new here and to using the openVibe. I’ve been attempting to get the EOG Denoising Calibration and EOG Denoising boxes going in the Designer. I haven’t had any problems actually getting them to work, but interpreting/making sense of the output has proven to be really confusing. It would be great if I someone here could have a look and let me know, if what I’m doing is right or wrong. I’m using openVibe 2.1 on a Windows 10 machine with a Muse Headband (TP9, AF7, AF8, TP10, Right AUX) that is streaming data to the openVibe acquisition server (I’m using the LSL driver marked unstable) over an LSL Bridge (BlueMuse).

To acquire the EEG signal I first connected the Acquisition Client to the Channel Selector, The channel selector was set to acquire/select the first 4 channels (TP9, AF7, AF8, TP10) of the Muse headband. There were two sessions of this I ran, the first one recorded a “clean” signal which involved sitting still. The second session to obtain the EOG artefacts in the signal was carried out by blinking almost throughout the signal acquisition phase and varying the type of blinks throughout.


For the calibration phase, I modified the existing scenario that is available with the openVibe in the following manner. I have used two GDF file readers to input the EEG and EOG signals. The EEG signals used is the “clean” signal I’ve recorded, while the EOG signal is the one with a large amount of eye blinks. On running this scenario openVibe generates the b-Matrix as it should.

Image


I have then used the EOG Denoising box in the run scenario and modified the scenarios in the manner you see in the image below. You’ll see that I have selected the EEG signal that goes into the box instead of deselecting it. Deselecting it results in an error since no signals are then being sent to the EOG Denoising box. Unfortunately, when I use the “clean” EEG signal for the EEG input along with noisy EOG signal for EOG input my processed signal looks exactly like my “Resting EEG” signal. Is this what it is supposed to look like? And is the box doing what it’s supposed to? Or has the manner in which I’ve connected the boxes and/or implemented the denoising calibration resulted in something I’m not able to understand? Here’s a screenshot of the all three sinals: Resting EEG, EOG and the Processed Signal
Image

Is it possible that the excessive blinking that can be seen in the EOG signal has somehow resulted in the b-Matrix that results from the calibration process not being able to pick up the occasional blink resulting in identical EEG and Processed Signal outputs?

If I follow the entire procedure without selecting channel 1 – 4 during the signal acquisition process, and follow the scenario examples exactly as they are (only changing the EEG, EOG and bMatrix files) and reject channels 1-4, then I end up with only the Right AUX signal as the end product, post EOG Denoising. This has me a bit confused, hence the long post.

Your assistance is greatly appreciated.

Thanks,

Amit.

jtlindgren
Posts: 775
Joined: Tue Dec 04, 2012 3:53 pm
Location: INRIA Rennes, FRANCE

Re: EOG Denoising

Post by jtlindgren »

Hello Amit, welcome on board!

I'm not very familiar with the boxes in question, and their manual pages are unfortunately not the most clear ones. Reasoning about the example scenarios, especially the channel selector configs in them, suggest that the box might actually not be intended to work in a way that two different recordings are sent to the two different inputs. This is indeed confusing as from machine learning you might well expect it to want signals or examples from different categories. If it was the blink/noblink machine learning 2 class dicothomy, why would the denoising box also have two inputs, and not just the calibration box?

My academic guess is that whats going on is this: the method expects data from a single recording session, where some electrodes are more contaminated than the others. These would be the frontal electrodes, or some custom electrodes attached very close to the eye, for example. It is the output of these custom electrodes that I guess should go to the EOG socket, and the rest of the electrodes to the EEG socket.

Does this help?

You could also find+read the Schlogl 2007 article the method is based on, that should confirm my suspicion (or not).

(edit: moved the thread to a more appropriate subforum as its about box behavior...)

Best,
Jussi

amitpbarde
Posts: 7
Joined: Wed Sep 12, 2018 4:56 am

Re: EOG Denoising

Post by amitpbarde »

Hi Jussi,

I really appreciate the prompt response. If I'm understanding what you've said, I should be using the electrodes that register the most EOG activity as the 'triggers' to run the calibration i.e. use the channel selection boxes exactly the way the scenario demonstrates their use. I will put this to the test and compare the results. Once again, thank you for your assistance.

Amit.

Post Reply