OBJECTIVES: Brain-computer interfaces that activate exoskeletons based on decoded movement-related activity have been shown to be useful for stroke rehabilitation. With the advances in the development of exoskeletons it is possible to replicate a number of different functional movements that are relevant to rehabilitate after stroke. In this study, the aim is to detect and classify six different movement tasks of the lower extremities that are used in the activities of daily living.
APPROACH: Thirteen healthy subjects performed six movement tasks 1) Stand-to-sit, 2) Sit-to-stand, 3) Walking, 4) Step up, 5) Side step, and 6) Back step. Each movement task was performed 50 times while continuous EEG was recorded. The continuous EEG was divided into epochs containing the movement intention associated with the movements, and idle activity was obtained from recordings during rest. Temporal, spectral and template matching features were extracted from the EEG channels covering the motor cortex and classified using Random Forest in two ways: 1) movement intention vs. idle activity (estimate of movement intention detection), and 2) classification of movement types.
RESULTS: The classification accuracies associated with movement intention detection were in the range of 80-90%, while 54±3% of all movement types were classified correctly. The stand-to-sit and sit-to-stand tasks were easiest to classify, while step up often was classified as walking.
SIGNIFICANCE: The results indicate that it is possible to detect and classify functional movements of the lower extremities from single-trial EEG. This may be implemented in a brain-computer interface that can control an exoskeleton and be used for neurorehabilitation.