Detection and Classification of Three-Class Initial Dips Using Vector Phase Analysis with Dual Threshold Circles: An fNIRS Study

In this paper, we have used dual-threshold circles in the vector phase analysis to maximize classification accuracy of three class initial dip in functional near-infrared spectroscopy (fNIRS)-based brain-computer interface (BCI). fNIRS signals for three different mental tasks (mental arithmetic, mental counting and puzzle solving) are measured from prefrontal cortex of eight healthy subjects. After the acquisition of signals, filters are applied to remove different noises (cardiac, respiratory and Mayer). A dual threshold-based vector phase analysis is used for the detection of initial dip. Features used for classification of initial dips are signal mean, signal slope, skewness, kurtosis and peak in three different window sizes 0~1, 0~1.5 and 0~2 sec. The features are then passed from multi-class linear discriminant analysis for classification of three different initial dips. The average classification accuracy in offline analysis obtained using dual threshold circles in the vector phase analysis is 69.12 % in the case of window size 0~2 sec. These results are a significant improvement compared to the existing literature and are more reliable for use in BCI applications.