Imagination detection right hand and left hand

About Brain Computer Interfaces, EEG, neurofeedback, related event announcements, jobs, etc.
Post Reply
FabioBCI
Posts: 7
Joined: Fri Nov 24, 2017 1:06 pm

Imagination detection right hand and left hand

Post by FabioBCI »

Hello everyone,
I have registered the movement imagination of the right and left hand with an 8-electrode g.Nautilus device, and I am trying to be able to detect the classes online.
The first thing I did was to record the data using a protocol where the task to be performed would be shown randomly and the user for 10 seconds imagined to move the right or left hand, the electrodes used to position them in C3, Cz and C4, well done an offline scenario to train an LDA classifier and the results are good, in the order of 70%, but when I use that classifier to classify new data, it gives me pure chance, I do not know what I'm doing wrong.
Attached my two scenarios the first is the offline treatment and the second to process the new data in a pseudo online.

thank you very much.
Attachments
offline_motor_imagery.xml
Offline process.
(77.2 KiB) Downloaded 237 times
pseudo_online_motor_imagery.xml
Pseudo online process.
(56.96 KiB) Downloaded 241 times

jtlindgren
Posts: 775
Joined: Tue Dec 04, 2012 3:53 pm
Location: INRIA Rennes, FRANCE

Re: Imagination detection right hand and left hand

Post by jtlindgren »

Hello Fabio,

I'm not sure if the electrode configuration is optimal for motor imagery. Do you have any other electrodes near C3 and C4? It'd be best if you could have some electrodes also around C3 and C4 for CSP to be able to build a Laplacian type center-surround configurations to better isolate the relevant sources.

The difference between train and online could be due to overfitting or nonstationarity or the subject being distracted by the results. One way to test if its the last one is to do two training runs (no blue bar feedback to user) sequentially, and then train classifier with the first run and test it with the second. I've noticed that for several users this gave much better accuracies that what they had if they got feedback.


Cheers,
Jussi

Post Reply