Optimizing Spatial Filters by Minimizing Within-Class Dissimilarities in Electroencephalogram-Based Brain–Computer Interface

M. Arvaneh, Cuntai Guan, Kai Keng Ang, Chai Quek
2013 IEEE Transactions on Neural Networks and Learning Systems  
A major challenge in EEG-based brain-computer interfaces (BCIs) is the inherent non-stationarities in the EEG data. Variations of the signal properties from intra and inter sessions often lead to deteriorated BCI performances as features extracted by methods such as common spatial patterns (CSP) are not invariant against the changes. To extract features that are robust and invariant, this paper proposes a novel spatial filtering algorithm, called KLCSP. The CSP algorithm only considers the
more » ... considers the discrimination between the means of the classes, but does not consider within-class scatters information. In contrast, the proposed KLCSP algorithm simultaneously maximizes the discrimination between the class means, and minimizes the within-class dissimilarities measured by a new loss function based on the Kullback-Leibler (KL) divergence. The performance of the proposed KLCSP algorithm is evaluated on the publicly available BCI Competition III dataset IVa, and a large dataset from stroke patients performing neurorehabilitation. The results showed that the proposed KLCSP algorithm significantly outperformed the CSP algorithm in terms of the classification accuracy (p < 0.01) by reducing within-class variations resulting in more compact and separable features. Index Terms-Brain-computer interface, common spatial patterns, EEG, non-stationary.
doi:10.1109/tnnls.2013.2239310 pmid:24808381 fatcat:egwswtp7prbn3dgzsj7namsqrm