bci2000.org BBS

Dear all,

I have a question. I am using the following spatial filter, classifier matrix and Matrix Buffer obtaining good results in my motor imagery BCI.

What I do not quite understand is if my classifier filter is well defined. The idea is to imagine the movements of both hands to get the ball in cursor task to go up and the movement of both feet to get the ball to go down. Do I have to change the weights? Maybe something like:

input channel input element(bin) output channel weight
C3_OUT 10Hz 2 0.5 (instead of -1)
C4_OUT 10Hz 2 0.5 (instead of -1)
Cz_OUT 10Hz 2 -1

I also do not quite understand the Matrix Buffer conditions. I attach the some pictures of my conditions.

Any light on those issues will be highly appreciated.

thank you!

Alejandro

First, for simplicity, I would recommend changing your output channel from 2 to 1. Then in the buffer conditions you can take the contents from column 2 and put them in column 1 then delete column 2.

I think your intuition about changing the weights in your classifier filter is a good idea. The optimal classifier weights will not necessarily be these nice round numbers, but rather something you can maybe determine using machine-learning techniques. But that might not be necessary at this stage.

As for the Matrix Buffer conditions, each row corresponds to an independent buffer of data maintained by the program. Whenever the criteria in the expression are met, data are added to the buffer. Each buffer gets equal representation in the calculation of the mean and variance of your control signal. It is useful to have a different buffer for each trial type to ensure that each trial type gets equal representation. This is important if you have a task that has, even transiently, unequal numbers of trials per class. You wouldn't want the mean to be calculated from mostly up trials and only a few down trials, for example, because then the mean -- which effectively becomes the decision point for deciding between class types -- would be closer to "up" state and it would thus be more difficult to classify trials as "up".

-Chad