Classification of stance and swing gait states during treadmill walking from non-invasive scalp electroencephalographic (EEG) signals

In Contreras-Vidal and colleagues have shown the feasibility of inferring the linear and angular kinematics of treadmill walking from scalp EEG. Here, we apply a discrete approach to the same problem of decoding the human gait. By reducing the gait process to a mere succession of Stance and Swing phases for each foot, the average decoding accuracy reached 93.71%. This is sufficient to design a gait descriptor that relies only on this classification of two possible states for each foot over time as input, which could complement the model-based continuous decoding method that lacks in some aspects (foot placement at landing, weight acceptance, etc.). A final implementation of this method could be used in a powered exoskeleton to help impaired people regain walking capability.