Optimizing spatial filter pairs for EEG classification based on phase-synchronization

Brain-Computer Interfaces (BCI) aim at translating brain signals, typically ElectroEncephaloGraphy (EEG), into commands for external devices. Spatial filters are powerful tools for EEG classification, able to reduce spatial blurring effects. In particular, optimal spatial filters have been designed to classify EEG signals based on band power features. Unfortunately, there are other relevant EEG features for which no optimal spatial filter exists. This is the case for Phase Locking Value (PLV) features, which measure the synchronization between 2 EEG channels. Therefore, this paper proposes to create such a pair of optimal spatial filters for PLV-features. To do so, we optimized a functional measuring the discriminability of PLV-features based on a genetic algorithm. An evaluation of our algorithm on a motor imagery EEG data set showed that using optimized spatial filters led to higher classification performances, and that combining the resulting PLV features with traditional methods boosts the overall BCI performances.