Mu Cursor Task Configuration

[chapi]

Hi all,
I’m working on the cursor task application trying to correctly configure my parameters but I have many issues.
I’m using a 16 electrodes configuration, following the 10-20 convention; my amplifier is the v-amp (brain vision products http://www.brainproducts.com/productdetails.php?id=15); all the impedences are under 10KΩ (checked with ActiCAP Control).

After the offline analysis with the stimulus presentation, I found out the channels and the frequencies to set up the linear classifier: left hand channel C4 16Hz; right hand channels C3 16Hz + FC1 20 Hz; both hands channel Cz 16hz; feet channel Cz 28Hz.
About the filtering tab:
1. Should I list all the channels in the list of CAR output channels?
2. How should I set up the weights in the matrix classifier?
3. Which is the correct way to use 2 channels as the same input channel inside the matrix classifier?
About the source tab:
1. Which could be the proper value of the SourceChGain?
2. The gain must be the same for each channel?

Can anybody help me?
Thank you in advance

[mellinger]

Hi,

1. Should I list all the channels in the list of CAR output channels?

It is sufficient to list the channels that are actually used in the LinearClassifier matrix. However, it is easier to change configuration later on when all channels are available after the SpatialFilter. To achieve this, just leave the CAR output channels parameter empty, as described at
http://www.bci2000.org/wiki/index.php/U ... rCAROutput

2. How should I set up the weights in the matrix classifier?

You should begin with a configuration that uses only a single classifier row, as described in the mu rhythm tutorial at:
http://www.bci2000.org/wiki/index.php/U ... ier_Matrix
In this case, the absolute value of the weight is irrelevant (provided it is not zero) because the Normalizer filter automatically adjusts the control signal's gain after the LinearClassifier has been applied.

Once you have successfully established feedback from one signal at a time, you may begin with combining amplitudes from different frequencies and/or electrodes into a single output channel. In case of your analysis results, you might do a left vs right combination, using weight -1 for C4@16Hz, and +1 for C3@16Hz, instructing the subject to imagine left vs right hand movement rather than right hand vs rest, as you would do when giving feedback from C3@16Hz only. For both hands vs rest, you will probably get a better signal when combining C3@16Hz + C4@16 Hz than when using Cz@16Hz.

Once you have established control in one dimension, you might progress to two-dimensional cursor control. For this, you need brain signals that can be modulated independently, and combine them.
E.g., you could do x = C3@16Hz + C4@16Hz, y = Cz@28Hz. In this case, you would instruct the subject to move the cursor in horizontal direction by imagining movement of both hands vs rest, and in vertical direction by imagining feet movement vs rest.

I strongly advise against giving feedback from FC1@20Hz because modulated activity there is typically caused by facial muscles, i.e. it is an artifact that cannot be used for building a BCI.

3. Which is the correct way to use 2 channels as the same input channel inside the matrix classifier?

I'm not sure whether I understand your question. You may combine two or more input channels into a single output channel. Weights are typically chosen to be +1 or -1, and their signs depend on the relative contribution of a signal amplitude you want to achieve. It is then up to you to determine an appropriate instruction to the subject.

About the source tab:
1. Which could be the proper value of the SourceChGain?
2. The gain must be the same for each channel?

Set SignalType to float32, and all SourceChGain entries to 1.
Different gains may be specified for different channels but I wouldn't advise this because it might lead to confusion in data analysis.

More help about the vAmp source module is available at:
http://www.bci2000.org/wiki/index.php/C ... ns:vAmpADC

Regards,
Juergen

[chapi]

Hi Juergen, thank you very much for your help, it was very precious for my work.

I tried to follow what you suggested: so i started to configure my matrix classifier for left hand and then left hand vs. right hand.: C4 @18Hz weight +10, C3 @16Hz weight -15; i used the following buffer condition. The cursor movement toward left direction was good and fast but for the opposite direction was too slow, even if i increased the value of the weight the movement was the same slow. Then I used more frequencies for C3 in the matrix classifier:
C4 18Hz 1 10
C3 14Hz 1 -15
C3 16Hz 1 -15
C3 18Hz 1 -15
but nothing changed. Do you have any idea to solve this problem?

After I tried to define the matrix classifier for hands and feet with the following configuration:
C3 18Hz 2 -10
C4 18Hz 2 -10
Cz 28Hz 2 10

Why is there an error when I use C3+C4 in the same row?
After this trial i made the full matrix classifier in order to use the cursor task for the two directions movement but i still
didn't get good results, especially for the vertical cursor movement. Is it possible that the system makes confusion cause i use the same frequencies for C3 and C4 in both directions 1 and 2?
C4 18Hz 1 1
C3 18Hz 1 -1
C3 18Hz 2 -1
C4 18Hz 2 -1
Cz 28Hz 2 1

Best Regards and thank you again.
Roberto

[mellinger]

Hi Roberto,

The cursor movement toward left direction was good and fast but for the opposite direction was too slow, even if i increased the value of the weight the movement was the same slow.

you need to give the system consistent for a while, otherwise the Normalizer cannot adapt. Once it has adapted to the control signal's mean and gain, both directions should behave equally.
I suggest that you first use 1D cursor feedback as described in the tutorial. This way, you can make sure that you are using appropriate signals and instructions, and that the Normalizer is configured correctly. Once you have it working in 1D with all signals you are interested in, you may try switching to 2D control.

Why is there an error when I use C3+C4 in the same row?

I'm not sure whether I understand your question. In each row, you specify one input channel, and one output channel. Using multiple rows, you can combine multiple input channels into one output channel.

After this trial i made the full matrix classifier in order to use the cursor task for the two directions movement but i still
didn't get good results, especially for the vertical cursor movement.

You cannot expect good results in a 2D experiment unless you have trained your subject to successfully use each of the two control signals in 1D experiments.

Is it possible that the system makes confusion cause i use the same frequencies for C3 and C4 in both directions 1 and 2?

As long as your subject is able to control the sum of and the difference between C3 and C4 independently of each other, the system won't be confused. However, controlling two signals independently is quite difficult, and requires a lot of training. E.g., the subjects participating in this study:

Wolpaw und McFarland. Control of a two-dimensional movement signal by a noninvasive brain-computer interface in humans. Proceedings of the National Academy of Sciences of the United States of America (2004) vol. 101 (51) pp. 17849-54

had quite some BCI experience, yet for 2D control they needed additional training, which took from 22 to 68 sessions (each session lasted for 30min).

Best regards,
Juergen

[chapi]

Dear Juergen,
thanks for your fast and exhaustive answer, I understood I have to train the subject for a long time..anyway i will let you know.
I have just the last question about the matrix classifier:

I'm not sure whether I understand your question. In each row, you specify one input channel, and one output channel. Using multiple rows, you can combine multiple input channels into one output channel.

It is not possible to configure the matrix in the following way?


The error is the following:
2011-09-07T12:55:31 - LinearClassifier: Preflight: Channel specification in Classifier(4,input channel)=C4+C3 is invalid.

Regards
Roberto

[mellinger]

No, it is not possible to enter C3+C4 as an input channel. Only existing channels may be specified as input channels. To combine channels C3 and C4, you need to provide a row that specifies C3 as an input channel, and an additional row that specifies C4 as an input channel, and use the same output channel in both rows.

Regards,
Juergen