Optimal Channel Selection Based On Statistical Analysis In High Dimensional Nirs Data

Near-infrared spectroscopy (NIRS) is an optical imaging method that has recently been investigated for noninvasive Brain Computer Interfaces (BC!). The performance of NIRS-based BCI can deteriorate when the number of channels becomes larger. Here we present three types of channel selection methods based on ranked channels, pre-defined channel configurations and statistical analysis for high dimensional NIRS data. The optimal combination of channels is selected by the highest classification accuracy rate based on Linear Discriminant Analysis (LDA). Experimental results show that the three considered types of channel selection methods achieve higher classification performance by removing the noisy and non-informative channels. Also the proposed statistical channel selection method can reduce the computation time significantly without any loss of classification accuracy.