A new approach based on principal ERPs and LDA to improve P300 mind spellers

Abstract

Visual P300 mind speller is a brain-computer interface allowing an individual to type through his mind. To this aim, the subject sits in front of a screen full of letters, and when his desired one flashes, there will be a P300 response (a positive deflection nearly 300ms after stimulus) in his brain signals. Due to the very low signal-to-noise (SNR) of the P300 in the background activities of the brain, detection of this component is challenging. Principal ERP reduction (pERP-RED) is a newly developed method that effectively extracts the underlying templates of event-related potentials (ERPs) by employing a three-step spatial filtering procedure. In this research, we investigate the performance of pERP-RED in conjunction with linear discriminant analysis (LDA) to classify P300 data. The proposed method is examined on a real P300 dataset and compared to the state-of-the-art LDA and support vector machines. The results demonstrate that the proposed method achieves higher classification accuracy in low SNRs and low numbers of training data.