F. Sandoval et al. (Eds.): IWANN 2007, LNCS 4507, pp. 956–963, 2007. © Springer-Verlag Berlin Heidelberg 2007 Use of ANNs as Classifiers for Selective Attention Brain-Computer Interfaces Miguel Ángel López 1 , Héctor Pomares 1 , Miguel Damas 1 , Eduardo Madrid 2 , Alberto Prieto 1 , Francisco Pelayo 1 , and Eva María de la Plaza Hernández 1 Department of Computer Architecture and Computer Technology University of Granada {malopez,hpomares,mdamas}@atc.ugr.es, {aprieto,fpelayo}@ugr.es 2 Department of Experimental Psychology and Physiology of Behavior of University of Granada emadrid@ugr.es Abstract. Selective attention to visual-spatial stimuli causes decrements of power in alpha band and increments in beta. For steady-state visual evoked potentials (SSVEP) selective attention affects electroencephalogram (EEG) recordings, modulating the power in the range 8-27 Hz. The same behaviour can be seen for auditory stimuli as well, although for auditory steady-state response (ASSR), it is not fully confirmed yet. The design of selective attention based brain-computer interfaces (BCIs) has two major advantages: First, no much training is needed. Second, if properly designed, a steady-state response corresponding to spectral peaks can be elicited, easy to filter and classify. In this paper we study the behaviour of ANNs as classifiers for a selective attention to auditory stimuli based BCI system. Keywords: Artificial Neural Networks, brain-computer interfaces, selective attention, Auditory Steady-state Response. 1 Introduction Many types of BCIs have been developed based on the classification of different features extracted from EEG recordings. For example, BCIs based on Event-related brain potentials (ERPs) are ones of the most popular. ERPs are as indicators of brain activities that occur in preparation for, or in response to, discrete events [1]. The P300 is an ERP with a typical latency exceeding 300 ms that shows up after the stimulus is presented and a cognitive task, typically counting appeared target stimuli, is performed. One of the reasons for using the P300 in BCIs systems is because it is a large ERP with maximum amplitude in the range of units of microvolts, big enough to be detected even in single-trial experiments [2]. Other BCIs are based on the voluntary modulation by the subject of spectral bands, such as alpha (8-13 Hz), beta (14-20) Hz or theta (5-8 Hz). One of the first BCIs used the spectral power of alpha band as feature to extract and classify, based on the assumption that human beings can