Emotiv Epoc for BCI

[tomsons22]

Hi,

I've benn working with a group of people developing brain computer interfaces (BCI) using the Emotiv Epoc neuroheadset. Until now we have been working on two different BCIs, one speller based on the P300 signal, and one based on SSVEP (steady state visually evoked potentials). These two BCIs work fine, they are kind of accurate. Now we want to design a BCI that doesn't need external stimuli, based on motor imaginery. We began recently with this project, but in this case we are not getting the results we expected (although we still have a few things to improve). With high density EEGs people have developed accurate BCIs based on motor imaginery. These EEG have higher spatial resolution (more electrodes), higher sampling frequency and provide a higher signal to noise ratio. I wanted to know what other people in the subject think. Do you believe that resolution of the signals that can be obtained with the Emotiv Epoc is not enough to develop these BCIs? Or we have to improve the processing stage and the clasiffication algorithm?

Thanks!
Tomas

[CPG]

Do you mean you want to train the BCI based on the subject thinking about a particular motor pattern?

 

It sounds like your question is essentially about the detection and decoding of neuronal signals. To calculate the physical limits of detection you need a lot of information about the sensitivity, placement, sampling frequency, etc. of your electrodes. Then we need to know what algorithms are being employed to decode the signals in a meaningful way. (I'm guessing some of these specifications are available at some estimation from the manufacturer, while others may not be. For example, they might not specify regarding proprietary features of their technology).

 

Some motor outputs may have very broad and easily detectable signals, but others might be harder to detect and below the threshold of your detection, or far too variable (trial-to-trial // person-to-person) to be consistently trained and recognized by the BCI. I don't think we on this forum have enough information to tell you what you should be able to detect with your current machinery, but you can experimentally employ your system to find out the limits of these parameters for yourself. Then you would need to determine whether what you have is sufficient or if you need to improve your resolution, processing, etc.

 

Also, if you want the BCI to have more than one of these, you'd also want to be aware of potential overlaps in the computational "decoding", and make sure that the patterns you want to recognize are not only recognizable, but also distinguishable from all others.