some problems with offline analysis

[jane1985]

Hi, I am a new user of BCI2000. amplfier: g.USBamp. I follow "Mu Rhythm BCI Tutorial", but some changes as follows
channelnames: C1 C2 C3 C4 FC3 CP3 FC4 CP4
reference: left mastiod ground: right mastiod
sourcechoffset: 0 0 0 0 0 0 0 0
sourcechgain: 1 1 1 1 1 1 1 1
notch: 48~52hz
bandpass:0.5~60hz
signal type: float32
commonground:true
commonreference:true.
task: imagine left hand and right hand
my config is correct?
1 check impendence
from the BCI2000, the impendence is as follows
Amp 0 ch 01 (UA-2006.09.07/01): 59234.7 kOhms
Amp 0 ch 02 (UA-2006.09.07/02): 56070.2 kOhms
Amp 0 ch 03 (UA-2006.09.07/03): 63092.7 kOhms
Amp 0 ch 04 (UA-2006.09.07/04): 49319.1 kOhms
Amp 0 ch 05 (UA-2006.09.07/05): 45273.5 kOhms
Amp 0 ch 06 (UA-2006.09.07/06): 46190.4 kOhms
Amp 0 ch 07 (UA-2006.09.07/07): 44073.3 kOhms
Amp 0 ch 08 (UA-2006.09.07/08): 42543.2 kOhms
but from the matlab simulink, use gtec check impendence: 36 28 29 34 56 46 28 42(kohm).
why this two methods results is different? In BCI2000, how many the impendence is ok?
2 offline analysis problem

[jane1985]

topgraphy.jpg (22.75 KiB) Viewed 19548 times

spectra.jpg (27.5 KiB) Viewed 19548 times

r square.jpg (28.67 KiB) Viewed 19548 times

I use offline analysis tool.
Target Condition 1:states.StimulusCode == 1
Target Condition Label 1: left hand
Target Condition 1:states.StimulusCode == 2
Target Condition Label 2: right hand
the results are above figures. channel 5 is Fc3, channel 7 is Fc4.
I think the results are wrong. why only the right part has red color in the topgraphy?

[mellinger]

Hi,

your results look good. There seems to be a problem with low frequency noise, though. Make sure that all impedances are below 25kOhms.
You are right that right and left hand curves should be reverted between FP3 and FP4. However, you might get some distortion due to asymmetric choice of ground and reference. Use "CAR" as a spatial filtering option in OfflineAnalysis to avoid this.

Juergen

[jane1985]

Thank you very much.

during experiment, should I set Filtering/SpatiaFilter/SpatialFilter Type in the Configuration module? I didn't use any filter in this step.
but in the offline analysis, I have used the CAR filter to get the above figures. It seems ok in the r square figure. But I don't know whether the Topgraphy is right. I think both FC3 and FC4 location have the red color, but why only Fc4 have obvirous red color?

and why the impendence is difference between gtec simulink impendence module and BCI2000?

How to select the reference and ground? what's the meaning of asymmetric choice of ground and reference?

look forward to your reply, thanks

[mellinger]

during experiment, should I set Filtering/SpatiaFilter/SpatialFilter Type in the Configuration module? I didn't use any filter in this step.

Yes, you should select the CAR filter for online operation.

But I don't know whether the Topgraphy is right. I think both FC3 and FC4 location have the red color, but why only Fc4 have obvirous red color?

It is normal to have an asymmetry between left and right hemisphere.

and why the impendence is difference between gtec simulink impendence module and BCI2000?

If you have high impedances (as it is the case for you), then impedance readings will change all the time, no matter which program you use to read them out. It could also be that the gtec program switches the amplifier into a different state regarding ground/reference combinations, which could also affect the impedance reading.

How to select the reference and ground? what's the meaning of asymmetric choice of ground and reference?

Reference is the electrode against which all channels are measured. Thus, when your reference electrode is not centered but located on one side of the head, you cannot expect signals from one hemisphere to look like signals from the other hemisphere.

[jane1985]

thank you very much.
actually, above three figures are 1 run analysis.
I get 7 runs in totally.
the analysis of total 7 runs data as follow 3 figures.

both channel 3(C3) and channel 7(FC4) have high r square value in the two figures. I think it should be different. I don't know the reason. whether the result is correct? what's the indication of this result?

and the left hand &rest topgraphy is ok? I don't know how to explain the topgraphy，why most regions have red color?

[mellinger]

Both your amplitude difference and r^2 values are low, so don't try to interpret them too much. The overall red color in the amplitude difference plot is probably an extrapolation artifact.

Try to get your impedances below 25kOhms to improve your results.

[jane1985]

I use offline analysis tool.
Target Condition 1:states.StimulusCode == 1
Target Condition Label 1: left hand
Target Condition 1:states.StimulusCode == 2
Target Condition Label 2: right hand
the results are above figures. channel 5 is Fc3, channel 7 is Fc4.

Hi, from the spectra , the spectra value of left hand is very close to right hand, their curves are same. what's the problem with the curve? I think the curve indicates that left hand and right hand cannot be discriminated. Is my view right? why you think the result seems good?

[mellinger]

Hi, from the spectra , the spectra value of left hand is very close to right hand, their curves are same.

No, there is a visible difference between the curves. It would be better visible if the zero peak in the spectrum were not there, it leads to a rescaling of the plot. From the r^2 plots, one can tell that the difference is significant at 10 Hz.

what's the problem with the curve? I think the curve indicates that left hand and right hand cannot be discriminated. Is my view right? why you think the result seems good?

The r^2 plots show that left and right hand can be discriminated well. 0.4 is a relatively high r^2 value. The only problem is that, as far as one can see from the plots, the curve at FC3 looks much the same as it does at FC4. Typically, you would expect the "left hand" curve and the "right hand" curve to exchange roles when going from the left to the right hemisphere.

[jane1985]

6th trail left vs rest R2.jpg (28.14 KiB) Viewed 19510 times

6th trail left vs rest spectral.jpg (18.15 KiB) Viewed 19510 times

from the left vs rest r square figure, channel 4 (C4) around 26HZ has max R square, but from the spectra figure, the max spectra is around 10 HZ;

[jane1985]

JIMMY RIGHT1 trail R2.jpg (35.3 KiB) Viewed 19510 times

JIMMY RIGHT1 trail spectra.jpg (23.85 KiB) Viewed 19510 times

from the rest vs right R square figure, channel 6(CP3) around 22 Hz has max R square, but from the spectra figure, around 10Hz has max spectra value;

why the frequencies of R square and max spectra are not the same?

and from the above left vs right hand results which I posted several days before, the frequencies of max r square and max pectral are 10Hz, and the channels are FC3 and FC4. Why the results are not consistent with the " rest vs right hand" and " rest vs left hand"? how to determine the channel and best frequecy to do online experiment?

[mellinger]

why the frequencies of R square and max spectra are not the same?

r^2 values measure how well discrimination between two conditions is possible. Often, high amplitudes and large amplitude differences coincide with high r^2 values, but this need not be the case. If there are large amplitudes/differences and low r^2 values, this is typically due to noise at those frequencies. From your plots, it appears that you have a lot of low frequency noise. To get rid of that noise,
1) try to get your impedances below 25kOhms,
2) use a high pass filter in your amplifier.

Code: Select all

Why the results are not consistent with the " rest vs right hand" and " rest vs left hand"?

Because they are for different conditions. r^2 values contain information about noise as well as the signal, and for different conditions, noise will be different. You may expect amplitude spectra to be more consistent across conditions than r^2 spectra.

how to determine the channel and best frequecy to do online experiment?

Go for the highest r^2 value over all conditions. In the online experiment, use the condition that discriminates best, i.e. has the highest r^2 value, and the associated channel and frequency.