EEG-based Brain-Computer Interfaces for neurorehabilitation and control

The research fieldof this dissertation are noninvasive, electroencephalography (EEG)-based, Brain-Computer Interfaces (BCIs), and their use for neurorehabilitation and control purposes.Brain-Computer Interfaces are systems enabling a straightforwardcommunication between the brain and the outside, by recording the neural activity and directly translating it into control signals for a particulardevice(likea robotic arm, a machine, or a computer).Given their independence from thenaturalneuromuscularpathways, BCIs are envisioned as tools to restore communication and control in patients with severe motor impairments. Also, BCIshave recently emerged inneurorehabilitation, where they are employed to objectify thedesiredmodulations of the neural features, toguidethe practiceand boost rehabilitation.This dissertation includes several research activities fromthe two above-mentioned contexts. Each studybuilds up on the advancementsof the previous researchand introducesa furtherstep, either by investigatingnew configurationsof theavailabletechnology(chapter 2), introducing some novel design elements(chapter 3), contributing into the practical implementation of new approaches(chapter 4), or improving the efficiency of available algorithms (chapter 5).The work is organized into five chapters. Chapter 1serves as introduction toEEG-based BCIsand their use inneurorehabilitation and control. In chapter 2, a novel combination of two neurorehabilitation tools is investigated,namelyi) BCI-guided motor imagery training and ii) transcranial direct current stimulation (tDCS). The work in chapter 3falls within the context of co-adaptive BCIs based on the modulationof sensorimotor rhythms, suggestingsome novel elements to improve the flexibility and tailoring of user training. In chapter 4real-time, continuous control of a robotic arm by means of continuously EEG-decoded movements is enabled for the first time. Chapter 5finally closes the thesis, with proposing two simple but effective ways to significantly improve SSVEP recognition based on Canonical Correlation Analysis (CCA)