A different scenario for motor-imagery BCI

[ddvlamin]

Hi,

As I'm mostly working on motor-imagery BCI, I wanted to have a scenario to train a spatial filter just like with the XDAWN P300 scenario. Because the basic CSP algorithm performs great for this type of task and it is easy to implement, I tried to make a new scenario. For now the implementation of CSP is only for two classes and it is a trace normalized version (according to the implementation I found in the matlab toolbox of Fabian Lotte), which to my experience works really better than the unnormalized version. The scenario computes 6 different spatial filters, each in a different frequency band: 8-12Hz, 12-16Hz, 16-20Hz...

Maybe it was already being implementend (although I looked in different branches and could not find anything), in that case it was a good exercise. If it was not already being implemented, here you can find it and play with it: http://www.thewired.be/blog/wp-content/ ... envibe.zip

I tested it last week and it seemed to work, although one can never be sure after one test. So if you find bugs in the scenario or the code, please let me know.

placing the two source files in the same directory as the XDAWN algorithm and compiling it should do it (it is based on the XDAWN implementation).

Best regards,
Dieter

[lbonnet]

Hi Dieter !

Thanks for the contribution!
Indeed we don't have an official implementation of CSP in OpenViBE, but I agree with you, it's proven to be very efficient and it is definitely a must-have.
Firstly, I will look at your implementation and test it.

I keep you posted

Laurent

[yrenard]

Sounds promising, thank you Dieter for sharing this
Yann

[lbonnet]

Hi Dieter !

I tested your CSP Spatial Filter Trainer and the scenarios you provided.
First, warm thanks for this contribution ! nice job you did here...
The compilation worked nicely on my side (windows7/64b).
I got some nice results knowing its was my first try of hands motor imagery.

Anyway... this contribution is a good candidate for official integration !
If you agree, we can schedule it for the upcoming OpenViBE spring release

If so, here is what we will have to do.
For you :
- Documentation of the box. This page explains how to do it.
As you are not a registered user on the SVN forge, I guess I will just need your .dox-part file completed. I'll do the documentation build on my own.
For me :
- Code review, to respect our coding rules. Basically I need to check the naming, the log messages, the debug messages...
- Test and validation on all supported platforms.
- Integration in trunk.

As your box relies on it++ library (the most reasonable choice, as we did for the xDAWN box), the box is under GPL License.
The code would be placed in openvibe-plugins/signal-processing-gpl/trunc.

If you are interested let me know ! And if you have questions don't hesitate...

Best Regards
Laurent

[ddvlamin]

This would be great

For you :
- Documentation of the box. This page explains how to do it.
As you are not a registered user on the SVN forge, I guess I will just need your .dox-part file completed. I'll do the documentation build on my own.

I'll take a look what I can do and let you know.

Best regards,
Dieter Devlaminck

[ddvlamin]

Hi,

In between some experiments with other people, I've created the dox-part file, including a sample scenario picture. I generated the html page and it looked good.

You can find it in the zip file on http://www.thewired.be/blog/wp-content/ ... x_part.zip

Best regards,
Dieter Devlaminck

[lbonnet]

Thank you Dieter !

I did some minor corrections in your documentation, nothing to worry about.
Few more tests and we're good for integration !

Best regards,
Laurent