JMIR Publications

ABSTRACT

Background:

Electroencephalography (EEG) is a non-invasive Brain Computer Interface (BCI) technology that has shown potential in various healthcare applications such as epilepsy treatment, sleep disorder diagnosis, and stroke rehabilitation. Usually these applications require multi-channels EEG. However, multi-channel EEG headset setup process is time consuming. This may result in low patients’ acceptance despite BCI potential benefits.

Objective:

To investigate the number of appropriate electrodes, which could be crucial for successful applications of BCI in wearable devices.

Methods:

Motor Imagery (MI) classification system is used for our analysis. Different number of EEG channels was selected. EEG Multi-frequency features were extracted by Filter Bank (FB). Support Vector Machine (SVM) was used in classifying left and right hand opening/closing MI task.

Results:

The results showed that the group of nine electrodes gave high classification accuracy while requiring moderate set-up time, and hence is suggested as the minimal number of channels. Spherical spline interpolation (SSI) was also applied to investigate the feasibility of generating EEG signal from limited channels of EEG headset. The classification accuracies of the interpolated groups only, and the combined interpolated and collected group, were significantly lower than those of measured groups

Conclusions:

For wearable device, one of the key factors that need to be concerned is wearability. The number of channels of EEG device adversely affects to set-up time. With FB feature and session dependent training, the investigation of number of channels provides the possibility to develop a successful BCI application using minimal channels EEG device. Interpolation technique which could approximate additional electrode data from nearby electrodes should be also explored.