Investigation of initial dips in mental arithmetic tasks: An fNIRS study

Abstract

In this paper, we investigate the feasibility of identifying the functional near-infrared spectroscopy (fNIRS) signal occurred from a single trial arithmetic task, in which the rest state hemodynamic response (HR), the occurrence of an initial dip, and the regular hemodynamic response are involved. fNIRS signals are measured from five healthy subjects for mental arithmetic tasks from the prefrontal cortex. Multiclass linear discriminant analysis (LDA) is used in classifying the fNIRS signal upon a single trial. Four different features including the signal mean, skewness, signal slope, and kurtosis are compared with five different window sizes: 0∼1, 0∼1.5, 0∼2, 0∼2.5, and 0∼3 sec for classification. Threshold-based vector phase analysis method is used to ensure the presence of initial dips in fNIRS signals. The average classification accuracy in offline analysis of 65.3% in 0∼3 sec time window using signal mean and signal slope is obtained. The result shows that the initial dip can be classified from the baseline (rest) and HR by using signal mean and signal slope as a features. This will result in the reduction of time window size to 0∼3 sec in order to use fNIRS signals for brain-computer interface (BCI).