An automatic subject specific channel selection method for enhancing motor imagery classification in EEG-BCI using correlation

A motor imagery (MI) based brain-computer interface (BCI) decodes the motor intention from the electroencephalogram (EEG) of a subject and translates this into a control signal. These intentions are hence classified as different cognitive tasks, e.g. left and right hand movements. A challenge in developing a BCI is handling the high dimensionality of the data recorded from multichannel EEG signals which are highly subject-specific. Designing a portable BCI whilst minimizing EEG channel number is a challenge. To this end, this paper presents a method to reduce the channel count with the goal of reducing computational complexity whilst maintaining a sufficient level of accuracy, by utilising an automatic subject-specific channel selection method created using the Pearson correlation coefficient. This method computes the correlation between EEG signals and helps to select highly correlated EEG channels for a particular subject without compromising classification accuracy (CA). Common spatial patterns (CSP) are used to analyse imagined left and right hand movements and the method is evaluated on both BCI Competition III Dataset IIIa and right hand and foot imagined tasks on BCI Competition III Dataset IVa. For both datasets, a minimum number of EEG channels are identified with an average channel reduction of 65.45% whilst demonstrating an increase of >5% in CA using channel Cz as a reference.

Automated summary

A motor imagery based brain-computer interface decodes motor intentions from EEG signals to translate them into control signals. The paper presents a method to reduce the number of EEG channels while maintaining accuracy, utilizing an automatic subject-specific channel selection method based on Pearson correlation coefficient. This method helps to select highly correlated EEG channels for a particular subject without compromising classification accuracy.