Online multi-class brain-computer interface for detection and classification of lower limb movement intentions and kinetics for stroke rehabilitation

Abstract

Closed-loop BCIs have recently been proposed for neurorehabilitation. This concept can be extended to complex motor tasks by decoding the type of the attempted movement in a multi-class BCI. Therefore, the objective of this study was to detect movements from real-time EEG and classify two movement types associated with the movement kinetics. EEG traces corresponding to executed and imagined movements of 12 healthy subjects and attempted movements of six stroke patients were detected using a template matching technique. Signal segments were characterized by six features and classified with a support vector machine. The system correctly detected and classified 57 ± 3%, 53 ± 6%, and 47 ± 7% of the executed, imaginary, and attempted movements, respectively. The performance of the detector was ~80%, and the classifier performance was in the range 63–70%. There were on average 1.6–1.9 false positive detections per minute. The results indicate that it is possible to detect executed, imaginary, and attempted movements in real time and classify movement kinetics. With improved performance, this could potentially be combined with functional electrical stimulation to promote neurorehabilitation by providing meaningful afferent feedback according to the attempted task.