Reconstructing the cortical functional network during imagery tasks for boosting asynchronous BCI

Michel Besserve (1) , Jacques Martinerie (1) , Line Garnero (1)

1 NCIC - Neurosciences cognitives et imagerie cérébrale

Brain-Machine Interfaces are aimed at providing a new communication channel to link the human brain to external world without using the usual nervous and muscular pathways. The work undertaken in our lab is focused on the study and development of an asynchronous BMI: the user communicates trough self-paced imagined motor or cognitive tasks which are not triggered by an external stimulation system. The system is based on the real time analysis of electroencephalographic (EEG) recordings from scalp electrodes in a human subject. For this application, efficient quantification of EEG signals is required to identify the performed mental tasks. Currently used features are spectral power and coherence of respectively each sensors and each couple of sensors. In this paper, we propose to improve the spatial resolution of EEG data by using a preliminary reconstruction of the cortical activity. Power and coherence features are then computed at the cortical level to feed a Support Vector Machine classifier. We applied this new approach to BCI data recorded in our lab on 5 subjects. Our results show that reconstructing the underlying neuronal network dynamics improves the performance of the device compared to a usual sensor level approach.

Mots clés: Brain-computer interfaces; Neural Prostheses; Brain Imaging; Data analysis

Domaines: Neurosciences [q-bio.NC]; Neurosciences

Michel Besserve, Jacques Martinerie, Line Garnero. Reconstructing the cortical functional network during imagery tasks for boosting asynchronous BCI. Deuxième conférence française de Neurosciences Computationnelles, "Neurocomp08", Oct 2008, Marseille, France. ⟨hal-00331614⟩