Hi,
Is it possible to have left/right or both hands/feet Motor Imagery in OpenVibe by using bipolar electrodes: FC3-CP3, FCz-CPz, and FC4-CP4 that give 3 channels C3, Cz and C4 respectively? Since the system has only 3 channels, is it possible to have real-time BCI by using only 3 channels? Since this is the limitation of the system what can be the possible avenue?
Thanks in advance!
3 Channel MI BCI
Re: 3 Channel MI BCI
Hi,
You can try to run some motor imagery with only 3 channels, following the motor imagery exemples provided in OpenViBE.
Although it is possible, the low number of channels may lead to poor classification results.
Kind regards,
Thomas
You can try to run some motor imagery with only 3 channels, following the motor imagery exemples provided in OpenViBE.
Although it is possible, the low number of channels may lead to poor classification results.
Kind regards,
Thomas
-
motorimagery
- Posts: 3
- Joined: Wed Oct 25, 2023 2:12 pm
Re: 3 Channel MI BCI
But, do I use CSP at all then?
Also, how do I improve accuracy with 3 channels? is there any other pipeline? I get an accuracy of 60, but I plan to use asynchronously (without instructions). Do you have any advice for that?
Also, how do I improve accuracy with 3 channels? is there any other pipeline? I get an accuracy of 60, but I plan to use asynchronously (without instructions). Do you have any advice for that?
-
kyunghowon
- Posts: 11
- Joined: Tue Apr 04, 2023 9:14 am
Re: 3 Channel MI BCI
Hello.
CSP is optimized to maximize the variance of one class and minimize the variance of another class and vice versa.
In general, the first N and last N filters are selected. In your case, however, since there are only three channels, it cannot guarantee optimal performance. It would be similar without CSP in some cases.
Optimal or minimized channel selection is also one active research field, so it is worth exploring the current state of the art.
I think different machine learning-based algorithms also benefit from more number of channels, as a majority of algorithms utilize covariance matrices.
Moreover, for asynchronous BCI, you need to either make a two-stage classifier (on vs. off -> left vs. right) or a 3-class classifier (off vs. left vs. right).
Basic CSP is optimized for binary class, so you need to apply different methods or design multi-stage classification.
Also, you need to consider if the three channels are enough to detect on/off states.
Best regards,
CSP is optimized to maximize the variance of one class and minimize the variance of another class and vice versa.
In general, the first N and last N filters are selected. In your case, however, since there are only three channels, it cannot guarantee optimal performance. It would be similar without CSP in some cases.
Optimal or minimized channel selection is also one active research field, so it is worth exploring the current state of the art.
I think different machine learning-based algorithms also benefit from more number of channels, as a majority of algorithms utilize covariance matrices.
Moreover, for asynchronous BCI, you need to either make a two-stage classifier (on vs. off -> left vs. right) or a 3-class classifier (off vs. left vs. right).
Basic CSP is optimized for binary class, so you need to apply different methods or design multi-stage classification.
Also, you need to consider if the three channels are enough to detect on/off states.
Best regards,
-
motorimagery
- Posts: 3
- Joined: Wed Oct 25, 2023 2:12 pm
Re: 3 Channel MI BCI
Sorry, I do not understand on/off. Can you please explain a bit more about it?
-
kyunghowon
- Posts: 11
- Joined: Tue Apr 04, 2023 9:14 am
Re: 3 Channel MI BCI
Hello.
Sorry for the unclear terms.
On/off in my reply meant the period when the user wants to use a BCI system.
You have mentioned asynchronous BCI. In asynchronous BCI, no stimulus onset indicates the task period, but it is determined by the user.
Therefore, you first need to determine when the user wants to use a BCI system.
Off: the user does not want to use a BCI system. The BCI system should be off.
On: the user wants to use a BCI system, then you need to classify either left vs. right, etc.
Otherwise, you can classify off (nothing/rest) vs. left vs. right.
Hopefully it will be more clear.
Best regards,
Sorry for the unclear terms.
On/off in my reply meant the period when the user wants to use a BCI system.
You have mentioned asynchronous BCI. In asynchronous BCI, no stimulus onset indicates the task period, but it is determined by the user.
Therefore, you first need to determine when the user wants to use a BCI system.
Off: the user does not want to use a BCI system. The BCI system should be off.
On: the user wants to use a BCI system, then you need to classify either left vs. right, etc.
Otherwise, you can classify off (nothing/rest) vs. left vs. right.
Hopefully it will be more clear.
Best regards,