Ganglion - missing eyes closed alpha / Electrode Problems? [resolved]
Hi everyone,
I am a beginner in BCI's and EEG data processing and I am currently working on a project to classify motor imagery. My problem is that the results of the classifications are not good and I think that the reason for this are signals that I record with the ganglion board. I don't want to go too deep into my program, it is a normal CSP and it works so far with test data from another project. I also tested motion execution, but again it didn't look good. I used multiple movements, but let’s just stay with left and right hand movement for the moment.
I used two different dry electrodes (TDE-212 and TDE-200) and placed them above the motor cortex. The reference and ground electrodes are placed on the earlobe. The headband I use is costum made, but it is in general just like the headband from the OpenBCI headband-kit. I use Python. Brainflow is used for the data acquisition. I received the TDE-200 electrodes around 6 month ago and used them like 3-4 times for a short amount of time. I got the TDE-212 electrodes like 1 month ago (gold and Ag/AgCl). Both electrodes are stored in a plastic bag and the bags are stored in a box. Therefore, I don’t think that electrode oxidation is the problem here. The small TDE-200’s are mat grey while the bigger TDE-212’s are shiny silvery at the tip and shiny bluish/greyish at the rest. I won’t use gel because I read somewhere in the forum that the coating wears off faster with it.
I used the OpenBCI GUI and found that I do not see any change in any of the possible frequency bands when I perform or imagine a movement. I tried to classifier “rest” vs “movement” but this results in pretty much the exact same features. I also placed the electrodes on O1, O2, Fp1 and Fp2 to detect eye blinks and only in Fp1 and Fp2 was eye blinking detected (but I think this is because the actual moving eyelid). I set the notch filter to 50 Hz in the GUI and set the bandpass filter to either 5-50 Hz or 7-13 Hz. When the electrodes are not touching anything, the data fluctuates in the 0.8-1.2 muV range. When I place the electrodes on my head, I get data in the 1-3 muV range (with and without imagining or performing motion). A colleague of mine is using the Cycton board and he gets around 30-40 muV, but it is a different board...
1.) Question: How much muV can I expect from an electrode on my head from the ganglion board?
I did a impedance measurement using the GUI and it doesn't look good. Most of the time I have around 200K to 600K Ohm in each electrode (except earlobes). I wouldn't mind if I had the Cyton board (1G Ohm input impedance), but I have the Ganglion board. According to another thread here in the forum [1], the input impedance is between 620K and 670K ohms. There is another thread where impedance problems (or no alpha band activity) are reported [2], but it never came to a conclusion. Here they wetted the electrodes to get better results, but I think this is not an viable option.
So I did a quick test where I pressed an electrode against my skin (not too hard). I cleaned the skin with alcohol and placed/tested the electrodes on multiple positions.
Hair: Thin/normal. 5mm to 20mm long. Not easy to clean the skin with alcohol because of hair, but it didn’t made much of a difference (near to zero).
Forehead: Area around Fpz. Top of Head: Area around Cz. Back of Head: Area around Oz.
Stable: Small wiggle or position change changes nearly nothing. Unstable: The opposite.
Impedance on reference channels (earlobe): 15-18K Ohm (stable)
Impedance when the electrode touches nothing: 479.5K or 603.5K Ohm (stable)
Reusable 5mm Gold-plated Spike Snap Electrode [TDE-212B]:
Forehead (clean): around 21K Ohm (stable)
Forehead (not clean): around 48K Ohm (stable)
Top of head: sometimes 200-300K Ohm, sometimes 479.5-603.5K Ohm (unstable)
Back of head: rarely 200-300K Ohm, most of the time 479.5-603.5K Ohm (unstable)
Arm (clean/not clean): 603.5K Ohm (stable)
Reusable 5mm Ag/AgCl Spike Snap Electrode [TDE-212]:
Forehead (clean): around 18K Ohm (stable)
Forehead (not clean): around 40K Ohm (stable)
Top of head: most of the time 200-300K Ohm, full range 90-500K Ohm (unstable)
Back of head: most of the time 200-300K Ohm, rarely 479.5-603.5K Ohm (unstable)
Arm (clean/not clean): 603.5K Ohm (stable)
Disposable / Reusable Dry EEG Electrode [TDE-200]:
Forehead (clean): around 16K Ohm (stable)
Forehead (not clean): around 19K Ohm (stable)
Top of head: 20-30K Ohm (stable)
Back of head: most of the time 18-40K Ohm, rarely 80-200K Ohm (medium stable)
Arm (clean/not clean): 603.5K Ohm (stable)
2.) Question: Are the results within an expected range?
The TDE-200 electrode gave the best results, but again I couldn’t see any activity in the alpha band from the OpenBCI GUI! I have not used wet electrodes yet, but that would be the next step for me. So beside skin preparation:
3.) Question: Is there anything that I miss or that I could to better?
4.) Question: Or is the general statement for the Ganglion Board that it is mainly good for cVEP (code based visual evoked, what MindAffect Team has been shown) and not for motor imagery/execution?
Side question about scale factor:
5.) Question: Unlike the Cyton documentation about the scaling factor, the Ganglion documentation does not explicitly state that Brainflow automatically scales the raw data. Do I have to do this manually for the Ganglion board or was it just forgotten to mention it in the documentation?
I thank you in advance for all the answers. I will try my best to provide any further informations.
Comments
Leppiz, hi.
If you are not seeing changes in alpha (at say O1 or P3) between eyes open / eyes closed, this could just be that your own alpha production is low. Some individuals have lower eyes closed alpha than others. I'd suggest trying a test with a couple friends to see if you notice they produce more alpha. Another factor could be the general ability of the individual to relax.
Regarding your MI detection, other threads have mentioned that signal quality and amplitude is always somewhat better with passive wet electrodes over the passive dry setup you are using. This is why MindAffect uses their headband with water moistened pads. They realized signal level with passive dry just was not sufficient.
Appreciate your impedance measurements, they seem normal. And the TDE-200 does produce somewhat better results because of the sharpness of the tips digging into the skin. (The other series from FRI have rounded tips.) Still, wet electrodes will improve things further. It may be easy to compromise by using a small water or saline moistened absorbent pad placed under the comb.
re: Ganglion scaling. Ganglion does not have scale adjustments like Cyton. Thus no adjustment is needed.
Regards, William
Then I will test the whole thing with wet electrodes and on other people, but I still find it very strange that not even movement execution is detected.
Do you mean that i can use the pad under the dry electrodes? I don't mind if the the coating is ok with that.
Really? Then why does it say at the bottom of the documentation that the data must be scaled (here)? Last update of this page was on 7/27/2021 so I assumed it was still something i have to do. I would assume that the GUI automatically scales the data, but I only use the GUI to check data quality. Data acquisiton is done with Python (Brainflow). Nevertheless the data in the GUI doesn't look good either... I will definitely test wet electrodes.
Yes, a small pad / sponge material underneath.
re: scaling, BrainFlow is already doing that for you.
Hallo again,
So I've run a few more tests now. I put a small moisted pad under the electrode. Furthermore, I have used conductive paste. Unfortunately, I had not found any wet electrodes. So everything is tested with dry electrodes. The impedances looked like this:
Reference Electrode (earlobe):
Not clean: 33-35K Ohm
Cleaned: 23-26K Ohm
Moisted Pad: 8K-10K Ohm
Conductive Paste: 2.5K-4K Ohm
Reusable 5mm Gold-plated Spike Snap Electrode [TDE-212B]:
Hair (not clean): 500K-600K Ohm
Hair (clean): 500K-600K Ohm
Hair (moisted pad): 13K-40K Ohm
Hair (conductive paste): 8K-13K Ohm
Reusable 5mm Ag/AgCl Spike Snap Electrode [TDE-212]:
Hair (not clean): 200K-500K Ohm
Hair (clean): 200K-500K Ohm
Hair (moisted pad): 13K-40K Ohm
Hair (conductive paste): 8K-13K Ohm
Disposable / Reusable Ag/AgCL Dry EEG Electrode [TDE-200]:
Hair (not clean): 20K-300K Ohm
Hair (clean): 20K-300K Ohm
Hair (moisted pad): 11K-14K Ohm
Hair (conductive paste): 5K-10K Ohm
Furthermore, I have implemented a P300 test. The electrodes ([TDE-212]) were placed at Fpz, Cz, Pz and Oz. Here a standard and odd tone was played. I then performed this once with bad impedances and once with good impedances (eyes open and eys closed). Result: I could clearly see the P300 (as well as other ERP's).
With confidence, I then performed another motor execution exercise. C4, Cz, C3 and C5 ([TDE-212]). Impedances were below 10K Ohms. Normal CSP was used. Movements: Shoulder (left and right), feet, hands (left and right), tongue and rest. One-Vs-One.
Result: Movements still cannot be classified well. Features in feature space are still mixed and not well separated. Even a general classification between rest phase and motion looks bad.
I find this a bit strange, since the P300 looks good. I maybe get the chance to test my program on a person where good motor execution/motor imagery is already confirmed, but I am not optimistic about the results. For now i would assume the Ganglion Board is not suitable for movement execution/imagery (atleat not plug and play), BUT i didn't tested any wet electrodes yet. Maybe i will do something else... something that is easier than motor execution/imagery.