Development of a Hybrid EEG-NIRS Brain-Computer Interface for Multiple Motor Tasks

Non-invasive Brain-Computer Interfaces (BCI) have demonstrated great promise for neuroprosthetics and assistive devices. Here we aim to investigate methods to combine Electroencephalography (EEG) and Near-infrared Spectroscopy (NIRS) in an asynchronous Sensory Motor rhythm (SMR)-based BCI. We attempted to classify 4 di erent executed and imagined movements: Right-arm, Left-arm, Right-hand, and Left-hand tasks. Previous studies demonstrated the bene t of EEG-NIRS combination, without processing the NIRS signal with online implementable methods for an asynchronous paradigm. Since normally the NIRS hemodynamic response shows a long delay, we investigated new features, involving slope indicators, in order to immediately detect changes in the signals. Moreover, Common Spatial Patterns (CSPs) have been applied to both EEG and NIRS signals. Fifteen healthy subjects took part in the experiments, and, because 25 trials per class were available, CSPs have been regularized with information from the entire population of participants and optimized using genetic algorithms. Di erent approaches have been investigated for feature extraction, classi cation, and signal association. The results showed that a hybrid EEG-NIRS approach enhances the performance of EEG or NIRS separately. Better performances are achieved for the motor execution paradigm, probably due to the subjects' inexperience in motor imagery, despite the small dataset available.