|dc.description.abstracteng||In recent years, brain-computer interface (BCI) has been developing as a communication and rehabilitation tool for patients with neurological diseases, such as stroke and spinal cord injury. This system decodes the user's intention from the brain signal, e.g. electroencephalography (EEG), and translates it into device commands. For either rehabilitation or communication system based on BCI, the latency of decoding algorithms is the fundamental factor which directly determines the system's performance. In this thesis, movement-related cortical potentials (MRCP) were investigated as the signal modality for real-time detection of motor intention from EEG. Aiming at a short-latency brain switch based on MRCP, a manifold leaning based method was developed, yielding a true positive rate > 80 % and a latency < 300 ms, which significantly outperformed previous methods. Based on this brain switch, two closed-loop BCI systems were implemented for rehabilitation and communication purpose, respectively. In the rehabilitation system, the brain switch was used to trigger a motorized ankle-foot orthosis, mimicking real movement of dorsiflexion, for natural afferent feedback. This closed-loop system was tested on healthy subjects, and neuroplasticity measured by transcranial magnetic stimulation was induced with only ~15-min intervention, indicating an efficient neuromodulation tool for rehabilitation. Further, the brain switch was extended to a multi-class BCI for communication purpose, by combining an electrotactile menu. The online testing on healthy subjects demonstrated its desirable properties of gaze-independence, high information transfer rate, and vast applicability. Finally a pre-clinical measurement was performed on spinal cord injured patients for investigation of the MRCP features, as a first step of potential application of the proposed BCI systems on this group of patients.
In conclusion, the main out-come of this thesis is a short-latency brain switch (Chapter 3) based on the investigation of MRCPs (Chapter 2), which was implemented in closed-loop BCI systems for rehabilitation and communication (Chapter 4), with potential application on patients with stroke and spinal cord injury (Chapter 5).||de