Enhancement of classification accuracy of a time-frequency approach for an EEG-based brain-computer interface

Objectives: The aim of this paper is to develop a new algorithm to enhance the performance of EEG-based brain-computer interface (BCI). Methods: We improved our time-frequency approach of classification of motor imagery (MI) tasks for BCI applications. The approach consists of Loplacian filtering, band-pass filtering and classification by correlation of time-frequency-spatial patterns. Results and Conclusions: Through off-line analysis of data collected during a "cursor control" experiment, we evaluated the capability of our new method to reveal major features of the EEG control for enhancement of MI classification accuracy. The pilot results in a human subject are promising, with an accuracy rate of 96.1%.