Classification and normalization in Cursor Feedback Task...

Forum for discussion on different user applications
Locked
aloplop
Posts: 41
Joined: 03 Sep 2008, 07:20

Classification and normalization in Cursor Feedback Task...

Post by aloplop » 02 Oct 2008, 13:37

Hi,

I have modificated the Cursor Feedback Task so that now I have got a target on the left side on the window and another one in the right side (width=5%). The ball cursor starts in the position 50 50 50 and it should move to the left or to the right.

Well, we have tried using the same parameters as the Mu tutorial, but they don´t work properly. We have used 2 bipolar channels (FC3-CP3 and FC4-CP4).

What modification could you recommend to me: classification matrix, normalization values or cursorfeedbacktask.cpp (which I have modified to move the cursor only in the X direction) ??

The same problem stands for the next:

In Signal Processing I have programmed a feature extraction filter using FFT (Welch method) to calculate the Power Spectral Density in mu and beta bands and I have the same problem because maybe a linear classifier is not the best one. The out_signal is a matrix 2x2 with [mu_power beta_power] of both bipolar channels.

What can I do? Thanks.

Álvaro.

mellinger
Posts: 1097
Joined: 12 Feb 2003, 11:06

Post by mellinger » 02 Oct 2008, 14:09

Did you get reasonable r^2 values for your bipolar channels when analyzing the initial session?

aloplop
Posts: 41
Joined: 03 Sep 2008, 07:20

r^2 values

Post by aloplop » 06 Oct 2008, 03:32

Hi,

well, what happened is that the ball only moved in one direction. But, if I moved one leg, it changed its direction. So:

1º) where can I obtain these r^2 values??


2º) What I have tried is to see the values of the control signal in CursorFeedbackTask.cpp through the App.Screen, but I don´t know how to use it correctly to move the cursor.
I mean, the method used in the original cursor task...should be used always or there is another way to control the cursor??

Thanks,

Álvaro.

mellinger
Posts: 1097
Joined: 12 Feb 2003, 11:06

Post by mellinger » 06 Oct 2008, 05:33

1º) where can I obtain these r^2 values??
http://www.bci2000.org/wiki/index.php/U ... al_Session

http://www.bci2000.org/wiki/index.php/U ... f_EEG_Data
2º) What I have tried is to see the values of the control signal in CursorFeedbackTask.cpp through the App.Screen, but I don´t know how to use it correctly to move the cursor.
I mean, the method used in the original cursor task...should be used always or there is another way to control the cursor??
In case you haven't already done so, it might be a good idea to work through the entire mu rhythm tutorial first (http://www.bci2000.org/wiki/index.php/U ... I_Tutorial). This will help you in making sure that every aspect of the standard BCI experiment is functional. Once this is assured, you can try varying elements of the system one-by-one.

When starting from a functional mu tutorial setup, and restricting cursor movement to the vertical direction, you should be able to control the cursor as intended. Restricting cursor movement to the horizontal direction, and additionally swapping references to output channels 1 and 2 in the LinearClassifier and Normalizer parameters, should work as well.

--Juergen

aloplop
Posts: 41
Joined: 03 Sep 2008, 07:20

Using the gUSBamp

Post by aloplop » 07 Oct 2008, 07:38

Hi,

I have used the g-Tec USB amp for the tutorial. I modifyed the stimulus parameters in order to see only just 2 arrows: one to the left and one to the right. Also, I used two filters: 2-30 Hz and 2-60 Hz with a Notch filter in 48-52 Hz.

The problem is that when I go to the Offline Analysis the peak near 10 Hz doesn´t appear (mu rhythm) and another below 5 Hz does. Also, the values for r^2 are very little (less than the ones in the tutorial) ~ 0.05 or less !

What can I do? Practise more? Use another filter or sampling rate?? I also tried the Amp with the cursor task modified to see a target on each side (left & right) but the results are bad also.

Thanks.

mellinger
Posts: 1097
Joined: 12 Feb 2003, 11:06

Post by mellinger » 07 Oct 2008, 09:03

Hi,

just a few questions regarding the quality of recording:
What montage are you using?
Which electrode is your reference electrode?
Where did you place the GND electrode?
What material are your electrodes made of? Are they all the same material?
What impedance do you measure?

When doing eye blinks/eye movements, are your getting artifacts that look like the ones shown in the tutorial? When biting teeth, are you getting muscular artifacts as illustrated in the tutorial? With eyes closed, are you seeing a 8-12Hz alpha rhythm at the occipital electrodes?

--Juergen

aloplop
Posts: 41
Joined: 03 Sep 2008, 07:20

gUSBamp recording problems

Post by aloplop » 07 Oct 2008, 11:17

Hi,

thanks, Juergen, for your quick response. Here are the answers:
What montage are you using?
I modified the eeg64.loc to another one which contained the four electrodes I use for recording: FC3, CP3, FC4 and CP4. (I erased all except these).
However it seems that if I introduce as montage the eeg64.loc the result is the same.
Which electrode is your reference electrode?
The one in mi left ear.
Where did you place the GND electrode?
In my right ear.
What material are your electrodes made of? Are they all the same material?
They are all gold electrodes.
What impedance do you measure?
We do this using a mdl file for Matlab which is installed with the amplifier.
At the beginning it measured almost 4 or 5 Kohms. Howewver, after an hour more or less the impedance were between 5 and 7 Kohms, so I think it was almost correct.
When doing eye blinks/eye movements, are your getting artifacts that look like the ones shown in the tutorial? When biting teeth, are you getting muscular artifacts as illustrated in the tutorial? With eyes closed, are you seeing a 8-12Hz alpha rhythm at the occipital electrodes?
Well if I blink, or move a leg or move my jaws I can see these artifacts in the BCI2000 window for signals. I haven´t tried the alpha ryhthm experiment with eyes closed, but the next time I use the gUSBamp I´ll try it.


------------------------------------------------------------------------------


With the Stimulus presentation I couldn´t get good results using the states.StimulusCode....(the same as in the tutorial)

Also when I load .dat files obtained using the Cursor Task I don´t know whether to use states.TargetCode (as condition) and states.Feedback (as trial change condition). Maybe here is the problem or I should use other parameters with the amp.

I thank all your interest. Regards,

Álvaro.

mellinger
Posts: 1097
Joined: 12 Feb 2003, 11:06

Post by mellinger » 07 Oct 2008, 13:00

the four electrodes I use for recording: FC3, CP3, FC4 and CP4
So, if I understand correctly, you have now four unipolar channels rather than two bipolar ones (as you mentioned in an earlier post), and all four channels are referenced to your left ear?

For the electrode positions you mention, you should get a signal variation when you imagine/perform hand movements rather than feet movements. Also, you should record from more electrodes to account for different source geometries (C3/C4 as a minimum, additionally C1/C2 and C5/C6 to allow for a Laplacian reference).

Please refer to http://www.bci2000.org/wiki/index.php/U ... m#Geometry to get an idea why using more electrodes makes sense, at least for the initial session.

Also when I load .dat files obtained using the Cursor Task I don´t know whether to use states.TargetCode (as condition) and states.Feedback (as trial change condition). Maybe here is the problem or I should use other parameters with the amp.
For the initial session, please use the StimulusPresentation setup. To analyze CursorTask data files, use

Code: Select all

states.TargetCode.*states.Feedback
for the condition, and

Code: Select all

states.Feedback==1
as the trial change condition.

Regards,
Juergen

Locked

Who is online

Users browsing this forum: No registered users and 1 guest