Convolutional Neural Network for Functional Near-Infrared Spectroscopy in Brain-Computer Interface

Functional near-infrared spectroscopy (fNIRS) is an emerging neuroimaging modality which can be utilized in braincomputer interface (BCI). In order to interpret the signal, important features for classification are extracted from hemodynamic signal. In the previous studies, various combinations of features have been reported to achieve higher classification accuracy. However, selecting the best combination of features depends on various factors. As we know, convolutional neural network (CNN) is able to learn and generalize features automatically. In this paper, we implemented CNN in order to extract fNIRS features and observed better performance over a conventional scheme combining signal mean and peak classified by support vector machine (SVM).