Electrode reduction using ICA and PCA in P300 Visual Speller Brain-Computer Interface system

Abstract

Brain-Computer Interface (BCI) research aims at developing systems helping disabled people hereafter called subjects. Due to the fact that technology underlying BCI is not yet mature enough and still having shortcomings for usage out of laboratory, these prevent their widespread application. These shortcomings are caused by limitations in functionality of BCI system tools and techniques. The motivation of this work was to develop efficient BCI techniques including signal processing, feature extraction, pattern recognition and classification to improve the performance of P300 Visual Speller BCI system. Data sets used in this paper were acquired using BCI2000's P300 Speller paradigm provided by BCI competitions. Primarily, in the processing phase time domain and spatial domain feature extraction were applied. Followed by classification phase where various linear and extended linear classifiers were utilized. One of the main achievements of this paper is applying Independent Component Analysis (ICA) or Principal Component Analysis (PCA) as spatial domain feature extraction for dimensionality and artifact reduction. Reducing electrodes to half its original size highly improved performance with linear classifiers and yet outperformed the results of BCI competition winners with extended linear classifiers.