Ramin Afrah - School of Advanced Technologies in Medicine, Medical Image & Signal Processing Research Center, Isfahan University of Medical Sciences Isfahan, Iran
Zahra Amini - School of Advanced Technologies in Medicine, Medical Image & Signal Processing Research Center, Isfahan University of Medical Sciences Isfahan, Iran
Rahele Kafieh - School of Advanced Technologies in Medicine, Medical Image & Signal Processing Research Center, Isfahan University of Medical Sciences Isfahan, Iran
Alireza Vard - School of Advanced Technologies in Medicine, Medical Image & Signal Processing Research Center, Isfahan University of Medical Sciences Isfahan, Iran

Brain Computer Interface (BCI) systems withlots of applications helping to contribute to make a chancefor people to communicate with environment directly bybrain signals without any muscular reaction. In P300-basedBCIs, detection of P300 component is the main challengeand machine learning algorithms are the most commonlyused methods on these systems. Despite the fact that in thepast few decades the use of classical pattern recognitionmethods for P300 detection is relatively common, thesemethods also have disadvantages, such as need to manualfeature selection, poor performance in nonlinearitymanagement and limited regularization parameters. On theother hand, with development of deep learning models inrecent years and their ability to solve a major part of theabove mentioned shortcomings, application of deep neuralnetworks has been suggested as a proper alternative toanalyze BCI systems. A suitable type of deep neuralnetwork for P300 BCIs is fully connected (FC) neuralnetwork which has good performance on nonlinearproblems and also in comparison with other types of deepneural networks has a low computational complexity.Thanks to mentioned advantages, in this paper a FC neuralnetwork is applied on dataset II of BCI competition III andobtained %95.35 accuracy to discriminate two classes (P300and NP300). In comparison with classic machine learningmethods and other types of deep neural networks, thisresult can be acceptable in aspects of simplicity andcomputational complexity.

Brain Computer Inteface (BCI), Deep Neural Network, P300 Speller