Estudio comparativo de métodos para el reconocimiento de potenciales relacionados a eventos P300 para una interfaz cerebro-computador

Abstract

In recent years, the Brain Computer Interfaces(BCI) have been highly studied, due to they allow to interact with the environment without the requirement to use the peripherical nervious system. Consequently, The appliaction of this, has been very useful in the rehabilitation engineering. However, the traslation of the user's intent through of Electroencephalography(EEG) is still a challenge for the scientific community, consequently, the stimulation that allow to evoke responses in patterns form, for that the system can recognizes them, is necessary. An experiment highly used corresponding to the Oddball paradigm, that through of visual stimulus, allow to evoke a positive deflection in the parieto-central cortex to the 300 ms, when the subject is interested in a specific stimuli between aleatory stimulation, known as P300 potential. The P300 have a problematic in his recognition that consist in a low signal to noise ratio, this generate that the extraction techniques be reason of interest. In the present work, a comparative study between five P300-recognition methods is preformed: two standard methods reported in literature: Mean-Amplitude-LDA(MA-LDA) and Stepwise-LDA(SWLDA), and three novel methods based in the Canonical Correlation Analysis(CCA): MA+CCA-LDA, CCA with Regularizad Logistic Regression and CCA with Multilayer Perceptron(MLP). The methods were validated in a available dataset, that consisted in a BCI-P300 system implemented in a Speller. Using as evaluation metrics: the classification percentage and the computational cost. Also a measurement protocol in healthy people was developed, to implement the BCI-P300 Speller in real time, at the simulation Lab of the Universidad Antonio Nariño, using the device of EEG acquisition g.Nautilus-32 PRO and the public software BCI 2000