spatial filter

[yannick]

Hi,
I need your help for the spatial filter coefficients. I have 23 channels Fz is the reference, and i want to use C3 and C4 for my output to classify right hand movement, left hand movement.
But i have some errors with spatial filter because i take the motor-imagery-bci-classifier-trainer scenario's spatial filter coefficients as consequence i have a bad matrix size errors.

Do you have an idea please?

[yrenard]

Dear Yannick,

what computation do you want to achieve ?

Yann

[yannick]

i want to use C3 and C4 for record right and left hand movement.
I have 19 electrods+ 4 (références eyes,mastoid), i use Fz like reference.
And i don't understand spatial filter.

Can you help me please?

[yrenard]

Dear yannick,

the spatial filter allows to linearly combine a number of inputs to another number of outputs. It is equivalent to mutiplying the input sample vector by a spatial filter matrix in order to get an output sample vector. For example, imagine you want to apply a laplacian spatial filter as done in the sample motor imagery scenarios, this is equivalent to have 4 times the center electrodes substracted by each of the 4 neighboor electrodes for each of the output channels. It could be represented as something like this :

Code: Select all

                                                      | in_left_center  |
                                                      | in_left1        |
                                                      | in_left2        |
                                                      | in_left3        |
| out1 |   =   | 4 -1 -1 -1 -1  0  0  0  0  0 |   *   | in_left4        |
| out2 |       | 0  0  0  0  0  4 -1 -1 -1 -1 |       | in_right_center |
                                                      | in_right1       |
                                                      | in_right2       |
                                                      | in_right3       |
                                                      | in_right4       |

If you don't understand this thing, please consider finding a linear algebra / signal processing book, it will explain this much better than I can.

Best regards,
Yann

[yannick]

Hi ,
i see your post but i need more informations about the number 4.
Why you've "4 times the center electrodes substracted by each of the 4 neighboor electrodes for each of the output channels"?

For me, you've your ROI and your electrods, if you need to record C3 and C4 for have the distinction of right and left hand movements, why you apply this spatial filter with the number 4?

thanks

[yrenard]

Dear Yannick,

this is for normalization. Since I substract 4 times each neighboor electrode, I have to add the center electrode 4 times.

Hope this helps,
Yann

[yannick]

Hi,

I would like to know if it necessary to use a spatial filter to classify signals?
In fact, in my eeg configuration i've only 23 electrods like that ( EOG+EOD not on this picture).

In this case, i've only C3 and C4 channels for record hands movements.

Do you think i need spatial filter ?

[yrenard]

Dear yannick,

you may want to use the neighboor electrodes T3, Cz, F3, P3 and Cz, T4, P4, F4 as we did in the motor imagery BCI. That will probably not be as good as what we proposed, but I still believe that would be better than simply using C3 and C4 alone.

However, what hardware are you using ? That does not look like any hardware we already support. Maybe you could contribute and send your driver for the next release.

Hope this helps,
Best regards,
Yann

[yannick]

Hi,

In fact i've tested this configuration but after two hours, i've no classification results. It's normal?
For the driver, i come back to you after.

thanks

[yrenard]

Dear yannick,

I can't tell you much more than what I told you already, maybe there is a problem with your scenario, maybe you are not trained enough, maybe there is a problem with the acquisition, it can be so many different things. You have to study your signals and find what's wrong with them. I understand this is challenging, but I fear I can't do much for you based on this short discussion.

Best regards,
Yann

[yannick]

In fact, i have recorded a eeg file with my eeg device, this file match with a motor task ( 10 mn), ths subject imagine hands movements.
After i load this file with ths acquisition client, and i used a spatial filter on these electrods. The file is played each 10 min for see test the classifier trainer. My question is, if you have a same signals sequence is played, it's possible to have a output classifier on the console? Like that you know if your scenario is good.

thanks

[yrenard]

Dear yannick,

sorry I don't understand what you want to do. Are you reading an EEG file back from the acquisition server ? If yes, where are your stimulations stored ? How do you sync this EEG with the internal OpenViBE clock ?

Yann