Elimination of bioelectrical source overlapping effects from the EEG measurements

The transmission of electric fields from a primary bioelectric source through biological tissue towards measurement sensors is known as the volume conduction. This phenomenon is widely exploited in many biosensors used for the measurement of bioelectromagnetism. One example of such sensors is an electroencephalograph (EEG) used for the recording and monitoring the bioelectrical activity of brain. Raw EEG scalp potentials are characterized by weak spatial resolution. Owing to the volume conduction there is an overlapping of the contribution to each electrode from neighboring bioelectrical sources. Due to described reasons, multichannel EEG recordings tend to provide an unclear image of the activity of brain. In this study the performance of algorithms based on spatial filtering in the task of eliminating source overlapping from EEG-based BCI system is examined and compared. The algorithms of choice are Common Average Reference (CAR), Surface Laplacian (SL), Common Spatial Pattern (CSP) and Filter Bank Common Spatial Pattern (FBCSP). Additionally, a new approach to feature selection and knowledge extraction from BCI models is proposed. For that purpose criterion based on the Gini variable importance evaluation, commonly performed by the Random Forests algorithm, is used. For the better performance evaluation of proposed approaches, their effectiveness is tested during the classification of the dataset IVa provided for the BCI Competition III organized by the Berlin Brain-Computer Interface group.