Feature extraction for BCIs based on electromagnetic source localization and Common Spatial Patterns

Brain-Computer Interfaces (BCIs) provide a way to communicate without movement and can offer significant clinical benefits. Electrical brain activity recorded using electroencephalography (EEG) can be automatically interpreted by supervised learning classifiers according to the descriptive features of the signal. This paper investigates the performance of novel feature extraction based on a signal source localization and Common Spatial Patterns (CSP) methods. The proposed approach was evaluated by the reference EEG dataset yielding an average classification accuracy of 74.6 % for a chosen group of subjects. It is shown that CSP feature extraction performs significantly better when applied to source components compared to features from the original sensor domain.