Hierarchical decoding of grasping commands from EEG

Brain-Computer Interfaces may present an intuitive way for motor impaired end users to operate assistive devices of daily life. Recent studies showed that complex kinematics like grasping can be successfully decoded from low frequency electroencephalogram. In this work we present a hierarchical method to asynchronously discriminate two different grasps often used in daily life actions (palmar, pincer) from a combined set of motor execution and motor intention. We compared sensorimotor rhythms based features and time features from the low frequency spectrum for best discrimination results. Our results show not only the principle feasibility of the proposed method with detection of asynchronous motor intention at rates of 80% accuracy and subsequent grasping discrimination over 60%, but also that low frequency time domain features provide a more consistent detection pattern. Although the basis of this results is still an off-line analysis we are confident that these results can be transferred to on-line use.