Motor task recording/offline analysis problems

[wmilewski]

Dear all,

I am a beginner both in terms of using BCI2000 and the BCI field itself. For some time now, I've been following your tutorials to obtain mu rhythm parameters in an initial session (recording myself) and analyse that data offline. I am able to perform each step of the analysis, only that the resulting feature maps seem (to me!) simply wrong, or entirely random, and look nothing like those from other mu rhythm-related projects. Given my lack of expertise, I can only ask you for any general thoughts and hints you have on this subject, including any basic setup steps I might have missed.

I'm using the Emotiv headset - I'm aware that the default electrode positioning does not cover the motor cortex, but it seems that tilting the headset backwards can give a better coverage of the desired area. Also, I'm highly inspired by the IpsiHand project (aac-rerc.psu.edu/wordpressmu/RESNA-SDC/2011/04/27/ipsihand-direct-recoupling-of-intention-and-movement-washington-university-in-st-louis), where they seem quite successful using Emotiv for mu rhythm analysis.
I use the Emotiv contribution module and Stimulus presentation to record myself performing left- and right-hand movements. For each session ( 5-15 runs, each 4-5 sequences) I've tried a different hand movement type (moving fingers, closing and opening my fist) and slightly different headset positioning, the results vary but usually:
1)very very few channels and/or frequencies stand out

Few channels and frequencies

2) or there is a pattern but in rather high frequencies, and the amplitude actually increases instead of decreasing during these events

Regular pattern of a few channels in rather high frequencies

3) sometimes I get a feature map that looks entirely random, with higher r2 values scattered along all frequencies and channels

My config parameters are as follows...

Config parameters

I'm using the Emotiv software to ensure that there is good connectivity between the electrodes and the scalp. But as I am recording, the signal seems to be rather noisy (not out of control, but with occasional spikes with strength up to that of an eye blink). Does it mean I should apply more filtering during the recording session? If so, how do I do this?
Also, 12 out of the 14 sensor pads are felt pads, whereas the remaining two pads are missing so I use cotton balls. Could this also impact the recorded signal?

My ultimate goal is to develop an application capable of distinguishing between 2-3 mental states of the user. But for now I would like to be able to perform an offline analysis and obtain results characteristic for performing (or imagining) a motor task.

Please assume I might be unaware of some principal mistakes I made and kindly point them all out..

Thank you,
Wojtek

[boulay]

Right-hand vs Left-hand should work, but some researchers have better success with right-hand vs "counting backwards from 100 in steps of 7". Be sure not to mouth the numbers or close your eyes as you count.

If you get >25 Hz activity increasing during your motor task, it's probably muscle artifacts creeping in on your signal. It is highly unlikely, but some claim it is possible, to get "gamma" activity in EEG. Gamma would also behave as you describe: increasing during the task relative to baseline.

I don't have an Emotiv headset, so I can't comment on the cotton balls or what the signal quality should be. However, can you record a few seconds of data with your eyes open (looking at something informative) and record a few seconds with your eyes closed? Calculate the spectra from the posterior and parietal channels for both conditions. There should be a huge increase around 10Hz with eyes closed. Post graphs of the raw data and the spectra if you can.

If other infrequent artifacts are a problem, you might have success removing them with ICA. EEGLAB is the easiest way to do that. If your artifacts appear to occur about once per second, check to see if they coincide with your pulse. If they do then you can try moving the headset around a little to get away from large blood vessels.

[wmilewski]

boulay,

Thank you for your reply. Sadly, I have just noticed today that my reply (which I thought I posted quite a few days ago) never appeared here.. Well, my bad.

Now, getting back to our conversation:
It is a good point that larger blood vessels can get in the way. Knowing this, and trying to reduce other artifacts, I think I can do better now when recording eeg data.

Over the last couple of days I have done more testing and focused on left hand vs. rest, because I really have to be able to detect any mu rhythm activity before taking it one step further and actually trying to distinguish between left and right hand movements. I have extended the session duration time significantly, to about 18 runs with 10 sequences each, which is approximately 15 minutes of data... and obtained somewhat better results:

The values from frequencies smaller than 10Hz and larger than 24Hz fade out, which is good. It also seems that the activity between 10 and 24 Hz decreases during motor movement. The only problem is that this is still very few channels and frequencies when you compare it to results like this one, from the IpsiHand project (also using Emotiv):

[...] can you record a few seconds of data with your eyes open (looking at something informative) and record a few seconds with your eyes closed? Calculate the spectra from the posterior and parietal channels for both conditions. There should be a huge increase around 10Hz with eyes closed. Post graphs of the raw data and the spectra if you can.

Like I said I'm more of a beginner here and I'm not sure how to record data without using the Stimulus Presentation in BCI2000.. to me, the only way of doing this seems to be displaying two different stimuli and closing my eyes during on of them, and looking at something during the other, and then use that data to calculate the spectra in the Offline Alasysis tool. But I'm sure there is a better way to do that?

On a related note - during the Stimulus Presentation, when a stimulus is displayed.. I understand that its associated stimulus code is automatically inserted into the data, but how much of that data (time-wise) is actually 'labelled' by this stimulus code? E.g. if the stimulus duration parameter is set to 3s, does it mean that all 3s following the stimulus are then used in Offline Analysis obtain characteristic features for that condition?
And does any part of the eeg recording pass as 'unlabelled' by any state code? Because from what I gather, all breaks between one stimulus and another represent the 'Rest' state. Am I correct here?