Classification of prefrontal and motor cortex initial dips for fNIRS-based-BCI

In this paper, we have classified the initial dips that are detected from the prefrontal and motor cortices using functional near-infrared spectroscopy (fNIRS) for brain-computer interface (BCI). The fNIRS data of mental arithmetic, mental counting, and right-hand finger tapping tasks are acquired from 5 healthy subjects. Vector phase analysis with a threshold circle (as a decision criterion) is used to detect the initial dips. Five different features including signal mean, signal slope, signal minimum value, kurtosis, and skewness in 0~1, 0~1.5, 0~2, and 0~2.5 sec windows are computed using oxyhemoglobin (HbO) signals. Linear discriminant analysis is used for the classification of the data. The average accuracy of 66.6% is obtained using signal mean and signal minimum value in 0~2.5 sec window. We used a conventional hemodynamic response to extract the signal mean and signal slope as features in 2~7 sec window for further validation of our results. LDA-based classification resulted in 73.2% accurate results for conventional hemodynamic response. The results seem significant for BCI using initial dip features.