Motor imagery detection with wavelet analysis for NIRS-based BCI

Near infrared spectroscopy (NIRS) is a non-invasive functional brain imaging device, which measures hemodynamic responses induced by brain activities. In this paper we collect NIRS signals associated with motor imagery and rest states, and analyze these signals with wavelet analysis. To our best knowledge, wavelet analysis method has not been used for hemodynamic response of motor imagery, although it is one of widely-used time-sequential signal analysis methods. In order to explore the usefulness of wavelet analysis, we extract features using various wavelet bases and then evaluate which features are more useful by cross-validation. Our empirical results clearly indicate that wavelet analysis is useful for obtaining meaningful features of the hemodynamic response, by achieving the averaged classification accuracy of about 86%. Among various wavelet bases used in our experiments, discrete Meyer wavelet function achieved the highest performance with classification accuracy of 93%.