Non-invasive classification of cortical activities for brain computer interface: A variable selection approach

We propose to carry out a classification method for electro-encepfialographic signals (EEG), using the activities of cortical sources estimated with an EEG inverse problem. To overcome the difficulties caused by the high number of sources (approximately 10000), we use a multivariate variable selection algorithm: the zero norm Support Vector Machine (L0-SVM). This technique allows to extract a small subset of sources, which are the most useful to allow for the discrimination of the mental states. The whole approach is applied to an asynchronous Brain Computer Interface (BCI) experiment from our lab. It outperforms a method based on the direct measurement of EEG electrodes' activities.

[1]  Vladimir Vapnik,et al.  Statistical learning theory , 1998 .

[2]  Bernhard Schölkopf,et al.  Use of the Zero-Norm with Linear Models and Kernel Methods , 2003, J. Mach. Learn. Res..

[3]  Richard M. Leahy,et al.  Electromagnetic brain mapping , 2001, IEEE Signal Process. Mag..

[4]  Isabelle Guyon,et al.  An Introduction to Variable and Feature Selection , 2003, J. Mach. Learn. Res..

[5]  Sylvain Baillet,et al.  A Bayesian approach to introducing anatomo-functional priors in the EEG/MEG inverse problem , 1997, IEEE Transactions on Biomedical Engineering.

[6]  J. J. Ermer,et al.  Rapidly recomputable EEG forward models for realistic head shapes. , 2001, Physics in medicine and biology.