bci2000.org BBS

Hello,

I'm the bigginer of BCI researches.
I'm interested in using BCI 2000, and doing some offline analysis with Mario program.
I getted some trials of hand motor imagery vs foot motor imagery EEG signals, and the program of Mario is working well.
The Mario's program indicate the r2 values of each electrodes and frequencies.

My question is how to calculate the r2 values between the two classes.

I know the definition of r values (it is something indicating correration). But if there is quite a few difference between two class, I think the correlation values becomes 1 (not 0).
In this graph, if there is a big difference, the r2 values become higher.

There are a lot of BCI papers using r2 values to indicate the difference between two classes, but there is no articles about how to calculate the r2 values.

So let me know about how to calcutate r2 value.

And if possible, please open the source code of Mario program (also other program), or make a manuals.
Because I have a difficulty to understand how the program works and how
this graph calculated.

Sincerely yours

Mitsuru,

The calculation of the r^2 values is described in a dissertation that you may download from:

http://www.bciresearch.org/tmp/thesis.pdf

You can also find a Matlab routine rsqu.m in \src\tools\analysis\P3 that you can use.

Thank you also for the suggestion to improve on the manuals. We are working on dramatically improved documentation at present.

Gerv

Thank you for the rapid reply

I read this document and my question about r2 value was solved.
Thank you very much

And if possible, I also need some explanation about AR model transfer method, because there is only reference from book.

Sincerely yours

Mitsuru

Mitsuru,

For the AR-based Maximum Entropy Method spectral analysis method that we use, we typically cite the book that you mention. I believe it is also described in Numerical Recipes in C.

Gerv

I will try to get the book.

Thank you very much

I'll be grateful if you give me a Link for the dissertation where the r^2 Method is discribed
the following doesn't Work
http://www.bciresearch.org/tmp/thesis.pdf

thnks

Larby,

The link works again.

Gerv

thanks

I know the original post was made several years ago but is it possible to still get access to the dissertation link which discusses the calculation of r^2 values?

Thanks
Prasheel

works again

Gerv

Thanks I was able to access the file.

I had some questions related to figure 5.2 on pg 182 in the document.

Are the time courses shown in figure B processed or averaged over several trials of stimulus and non-stimulus?

What is the exact method that you would use to generate the r squared time course in figure C from the data in figure B.

Once you have the r squared time courses for all of your different electrode positions how can you use this information as the input to a linear discriminant analysis model to build a linear classifier.

Any detailed references related to these topics would be greatly appreciated.

Thanks
Prasheel

Prasheel,

As described in the caption to that figure, waveforms were first averaged for 15 stimuli. Then, r^2 values were calculated for each time point between the distributions of voltages for target and non-target stimuli.

For a linear discriminant analysis, you would not use r^2 values as the input but rather the voltages directly. This is described in detail in recent articles by Krusienski, Sellers, and Wolpaw.


Gerv

Hello!
Can somebody help me with the thesis document where r^2 is described better?

The links that are posted here are not working any more.

Thank you so much

Eliana,

Try http://www.bci2000.org/tmp/thesis.pdf

Gerv

THANK YOU SO MUCH!!