Selecting relevant electrode positions for classification tasks based on the electro-encephalogram

The aim is to describe a general approach to determining important electrode positions when measured electro-encephalogram signals are used for classification. The approach is exemplified in the frame of the brain-computer interface, which crucially depends on the classification of different brain states. To classify two brain states, e.g. planning of movement of right and left index fingers, three different approaches are compared: classification using a physiologically motivated set of four electrodes, a set determined by principal component analysis and electrodes determined by spatial pattern analysis. Spatial pattern analysis enhances the classification rate significantly from 61.3 +/- 1.8% (with four electrodes) to 71.8 +/- 1.4%, whereas the classification rate using principal component analysis is significantly lower (65.2 +/- 1.4%). Most of the 61 electrodes used have no influence on the classification rate, so that, in future experiments, the setup can be simplified drastically to six to eight electrodes without loss of information.