Adaptive tracking of discriminative frequency components in electroencephalograms for a robust brain-computer interface

In an electroencephalogram (EEG)-based brain-computer interface (BCI), motor imagery has been successfully used as a communication strategy. Motor imagery causes detectable amplitude changes in certain frequency bands of EEGs, which are dubbed event-related desynchronization\synchronization. The frequency components that give effective discrimination between different types of motor imagery are subject specific and identification of these subject-specific discriminative frequency components (DFCs) is important for the accurate classification of motor imagery activities. In this paper, we propose a new method to estimate the DFC using the Fisher criterion and investigate the variability of these DFCs over multiple sessions of EEG recording. Observing the variability of DFC over sessions in the analysis, a new BCI approach called the Adaptively Weighted Spectral-Spatial Patterns (AWSSP) algorithm is proposed. AWSSP tracks the variation in DFC over time adaptively based on the deviation of discriminative weight values of frequency components. The classification performance of the proposed AWSSP is compared with a static BCI approach that employs fixed DFCs. In the offline and online experiments, AWSSP offers better classification performance than the static approach, emphasizing the significance of tracking the variability of DFCs in EEGs for developing robust motor imagery-based BCI systems. A study of the effect of feedback on the variation in DFCs is also performed in online experiments and it is found that the presence of visual feedback results in increased variation in DFCs.