Classification of motor imaginary based near-infrared spectroscopy signals

Abstract

Near Infrared spectroscopy (NIRS) is a brain imaging technique that measures hemodynamic activity in the human brain cortex with special wavelengths (infrared) in the light. the use of this technique in brain-computer interface (BCI) systems is increasing in terms of noninvasive and is not affected by electrical noise. With this increasing use, works become more important for high-accuracy NIRS based BCI systems. For a high-performance BCI system, the preprocessing, feature extraction and classification methods applied to BCI signals are important. For this purpose, in this study, we were studied 2-class (hand opening-closing) motor imaginary NIRS data set recorded 29 subjects. Firstly, change in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) concentrations were determined by applying the modified Beer-Lambert law to the data set. Then, features were extracted by Katz fractal dimension from pre-processed HbR and HbO. the extracted features were classified by k-nearest neighbors and then we calculated 74.10% and 71.10% mean classification accuracy (CA) for HbR and HbO, respectively. These values are 5.86% and %6.64 higher than the average 66.50% and 63.50% CAs calculated in the literature for HbR and HbO. These results indicate that proposed method is effective for this data set.