Motor Intention Recognition in EEG: In Pursuit of a Relevant Feature Set Pablo A. Iturralde 1 , Martín Patrone 1 , Federico Lecumberry 2 , and Alicia Fernández 2 1 Department of Physics, School of Engineering, UdelaR, Montevideo, Uruguay 2 Department of Electrical Engineering, School of Engineering, UdelaR, Montevideo, Uruguay {iturral,mpatrone,fefo,aliciaf}@fing.edu.uy Abstract. Brain-computer interfaces (BCIs) based on electroencephalo- grams (EEG) are a noninvasive and cheap alternative to get a commu- nication channel between brain and computers. Some of the main issues with EEG signals are its high dimensionality, high inter-user variance, and non-stationarity. In this work we present different approaches to deal with the high dimensionality of the data, finding relevant descriptors in EEG signals for motor intention recognition: first, a classical dimen- sionality reduction method using Diffusion Distance, second a technique based on spectral analysis of EEG channels associated with the frontal and prefrontal cortex, and third a projection over average signals. Per- formance analysis for different sets of features is done, showing that some of them are more robust to user variability. 1 Introduction In recent years, Brain-Computer Interfaces (BCIs) have become an active topic of research. Such interfaces could provide an alternative communication channel between humans and computers, replacing the normal output channel of nerves and muscles. As its main application, BCIs could be used by paralyzed patients or others suffering some type of motor impairment but who are cognitively intact as a means to interact with the environment. BCIs register brain states (relating to thoughts and intentions) as signals that are interpreted and translated into actions. The signal acquisition process is crit- ical to the performance of the whole system, and several technologies have been proposed to carry out such a task. Signal registering through electroencephalo- grams (EEG) are one of the most promising systems because of its noninvasive nature that allows for simpler and cheaper devices with almost no associated risks, as opposed to invasive technologies such as electrocorticographic (ECoG) signals which require medical procedures for its implantation. However, EEG sig- nals provide only a diffuse access to brain signals, since currents in the brain cor- tex are volume conducted through the skull before being sensed at the scalp. This means that more sophisticated processing and recognition systems are needed in order to obtain information about brain states from such signals [12,10]. L. Alvarez et al. (Eds.): CIARP 2012, LNCS 7441, pp. 551–558, 2012. c  Springer-Verlag Berlin Heidelberg 2012