Independent Component Analysis for Spatial Filtering and Feature Extraction in a Four-Task Brain-Computer Interface

Independent component analysis (ICA) is one statistical method closely related to the method called blind signal separation. In this paper, three typical ICA algorithms FastICA, Infomax and SOBI are utilized for spatial filtering and feature extraction in a four-task brain-computer interface (BCI) by decomposing EEG signals into independent sources. These algorithms are applied to five data sets recorded during motor imagery based BCI experiment and compared with well known algorithm common spatial pattern (CSP) in terms of classification performance. Averaged classification accuracies over the five data sets achieved by FastICA and Informax are better than or equal to that yielded by CSP algorithm, verifying the usefulness and feasibility of ICA methods for multi-task BCI application.