Application of Phase Space Reconstruction in a Few-Channel EEG-NIRS Bimodal Brain-Computer Interface System

Abstract

We developed a highly accurate, few-channel, bimodal electroencephalograph (EEG) and near-infrared spectroscopy (NIRS) brain-computer interface (BCI) system by developing new methods for signal processing and feature extraction. For data processing, we performed source analysis of EEG and NIRS signals to select the best channels from which to build a few-channel system. For EEG feature extraction, we used phase space reconstruction to convert EEG few-channel signals into multichannel signals, facilitating the extraction of EEG features by common spatial pattern. The Hurst exponent of the selected 10 channels constituted the extracted NIRS data feature. For pattern classification, we fused EEG and NIRS features together and used the support vector machine classification method. The average accuracy of bimodal EEG-NIRS was significantly higher than that of either EEG or NIRS as unimodal techniques.