Rosipal R., Rostakova Z., Porubcova N., Trejo L.J.

We present a novel method of tensor decomposition of EEG for precise measurement and real-time tracking of narrowband brain oscillations (NBO) for brain-computer interfaces (BCI). To determine NBOs associated with specific limb movements, we used mirror-box therapy, in which users view mirror images of one limb moving to alter the NBO associated with the movement of the contralateral limb. Unlike purely imaginary motion, mirror-box imagery is specific and easy for users to control. Tensor decomposition is a machine-learning algorithm that separates the NBOs present in a multi-channel EEG and provides a spectral-spatial signature for precise measurement of each NBO. Using the signature of a NBO we can track its activity in real time by back-projecting the signature on a live multichannel EEG recording. This enables continuous monitoring of the synchronization and desynchronization of selected NBOs and the construction of an elegant BCI protocol. We applied this approach to rehabilitating post-stroke patients using the BCI control of a robotic splint and a virtual reality avatar hand.