Alertness States Classification By SOM and LVQ Neural Networks

Several studies have been carried out, using various techniques, including neural networks, to discriminate vigilance states in humans from electroencephalographic (EEG) signals, but we are still far from results satisfactorily useable results. The work presented in this paper aims at improving this status with regards to 2 aspects. Firstly, we introduce an original procedure made of the association of two neural networks, a self organizing map (SOM) and a learning vector quantization (LVQ), that allows to automatically detect artefacted states and to separate the different levels of vigilance which is a major breakthrough in the field of vigilance. Lastly and more importantly, our study has been oriented toward real-worked situation and the resulting model can be easily implemented as a wearable device. It benefits from restricted computational and memory requirements and data access is very limited in time. Furthermore, some ongoing works demonstrate that this work should shortly results in the design and conception of a non invasive electronic wearable device.

Authors:

• K. Ben Khalifa
• M.H. Bédoui
• M. Dogui
• F. Alexandre

Keywords:

• Electroencephalogram interpretation
• artificialneural networks
• vigilance states
• hardware implementation

References:

[1] Billiard M. Le sommeil normal et pathologique: troubles du sommeil et
de l'éveil. Masson (Ed), 1994. PARIS
[2] Kohonen, T. Self-Organization Maps. (3rd edition 2001). Springer,
Berlin.
[3] Bougrain, L. and Alexandre, F. Unsupervised Connectionist
Algorithms for Clustering an environmental data set : a comparison.
Neurocomputing. 1999. vol 1-3. n┬░ 28. pp.177-189.
[4] Kohonen T., Hynninen J., Kangas J., and Laaksonen J.. SOM_PAK:
The Self-Organizing Map Program Package. Technical Report A31,
Helsinki University of Technology, Laboratory of Computer and
Information Science, FIN-02150 Espoo, Finland, 1996.
[5] Kohonen T., Hynninen J., Kangas J., and Laaksonen J. and Torkkola K.
LVQ_PAK: The Learning Vector Quantization Program Package.
Technical Report A30, Helsinki University of Technology, Laboratory
of Computer and Information Science, FIN-02150 Espoo, Finland,
1996.
[6] K. Ben Khalifa, M.H. Bédoui, L. Bougrain, R. Raychev, M. Dogui,
F. Alexandre, Analyse et classification des états de vigilance par
réseaux de neurones. Rapport de recherche ├á l'INRIA. RR-4714.
Janvier 2003.
[7] A. Vuckovic, V. Radivojevic, ACN Chen, D. Popovic, Automatic
recognition of alertness and drowsiness from EEG by an artificial
neural network, Medical Engineering & Physics 24 (5): pages 349-360,
2002.
[8] K. Ben Khalifa, B. Girau, M.H. Bedoui, R. Raychev, F. Alexandre.
Implantation basse-consommation de cartes auto-organisatrices pour la
détection de l'hypovigilance. RFIA, Toulouse janvier 2004