Neural Decoding of Robot-Assisted Gait During Rehabilitation After Stroke

Abstract

OBJECTIVE: Advancements in robot-assisted gait rehabilitation and brain-machine interfaces may enhance stroke physiotherapy by engaging patients while providing information about robot-induced cortical adaptations. We investigate the feasibility of decoding walking from brain activity in stroke survivors during therapy using a powered exoskeleton integrated with an electroencephalography-based brain-machine interface. DESIGN: The H2 powered exoskeleton was designed for overground gait training with actuated hip, knee, and ankle joints. It was integrated with active-electrode electroencephalography and evaluated in hemiparetic stroke survivors for 12 sessions per 4 wks. A continuous-time Kalman decoder operating on delta-band electroencephalography was designed to estimate gait kinematics. RESULTS: Five chronic stroke patients completed the study with improvements in walking distance and speed training for 4 wks, correlating with increased offline decoding accuracy. Accuracies of predicted joint angles improved with session and gait speed, suggesting an improved neural representation for gait, and the feasibility to design an electroencephalography-based brain-machine interface to monitor brain activity or control a rehabilitative exoskeleton. CONCLUSIONS: The Kalman decoder showed increased accuracies as the longitudinal training intervention progressed in the stroke participants. These results demonstrate the feasibility of studying changes in patterns of neuroelectric cortical activity during poststroke rehabilitation and represent the first step in developing a brain-machine interface for controlling powered exoskeletons.