Abstract

Brain-computer interfaces (BCIs) that read electroencephalography (EEG) give people a way to control external devices with brain activity alone and without surgery. Over the past two decades the most active medical use of this idea has been stroke rehabilitation, where a BCI can reward a patient for imagining or attempting a movement that the paralysed limb cannot yet perform. This paper reviews the field with that application in mind. We describe how an EEG-based BCI is built, from signal acquisition and artifact removal to feature extraction and classification, and we compare the three training approaches used most often after stroke: motor imagery, movement attempt and sensorimotor-rhythm neurofeedback. We then summarise other clinical uses, including communication for people who are locked in, detection of awareness in disorders of consciousness, and EEG- based seizure detection. Controlled studies and meta-analyses suggest that BCI-assisted therapy can improve upper-limb function, but the trials are small and the benefit differs a great deal from one person to the next. we conclude with the practical hurdles to clinical implementation and the pathways that appear most promising: hybrid systems, adaptive decoders and home training.