Page 1 of 1

Offline calculation of performance of the classifier for motor imagery

Posted: 16 Jul 2015, 07:30
by fanny
Hi everybody,

Currently, I am working about Cursor task with motor imagery. However, I didn't find a tool to calculate the automatic weight of the classifier and an offline calculation of the performance of it. So, I wonder if a tool like P300Classifier exists for Stimulus Presentation task with motor imagery?

Thank you in advance for your help,

Fanny

Re: Offline calculation of performance of the classifier for motor imagery

Posted: 16 Jul 2015, 09:30
by pbrunner
Fanny,

in my experience it works best for SMR experiments to initially limit the possible control channels to a laplacian filter over C3, Cz, or C4. You can perform an initial screening with actual and imagined opening/closing of the left and right hand. Then you use the OfflineAnalysis Tool in BCI2000 to identify which of the three possible control channels work best. For the SMR imagery training a basket task without a time-out works best. For this you first establish the mean and std of the signal using the adaptation (value=2). You then disable the adaptation (value=0) and have the subject train the imagination tasks.

I will send you slides that describe this process and the corresponding parameter files per e-mail.

Regards, Peter