Hybrid EEG-NIRS BCI using Feature Extraction based on Riemannian Geometry

Brain-computer interface (BCI), which enables direct control of computers and external devices through brain activity, holds significant promise for rehabilitating individuals with motor dysfunctions. However, problems such as low accuracy and non-stationarity of physiological signals remain. In this study, we propose a hybrid BCI using electroencephalography (EEG) and near-infrared spectroscopy (NIRS) with a feature extraction method based on Riemannian geometry. Our study analyzed EEG and NIRS data from 29 healthy subjects during motor imagery of left or right hand opening and closing movements. We discuss the effectiveness of the proposed method by discriminant performance.

I. Introduction

Brain-computer interface (BCI) enables direct control of external devices through brain activity evoked by motor imagery (MI) and external stimuli, facilitating communication without relying on traditional motor pathways. Therefore, we expect BCI to support daily life and rehabilitation of individuals with severe motor dysfunctions. However, due to individual differences and the non-stationarity of physiological signals handled by BCI, the accuracy of BCI is low, and it has not yet become widespread. Fazli et al. [1] proposed a hybrid BCI using electroencephalography (EEG) and near-infrared spectroscopy (NIRS). The results showed a mean accuracy improvement of 5% in more than 90% of subjects compared to EEG-only BCI. To address non-stationarity, Barachant et al. [2] proposed BCI using a method based on Riemannian geometry. Compared with conventional methods, Barachant’s method showed high robustness against the non-stationarity of EEG. In this study, we propose a hybrid EEG-NIRS BCI with a feature extraction method based on Riemannian geometry.