Directed neural connectivity changes in robot-assisted gait training: A partial Granger causality analysis

Abstract

Now-a-days robotic exoskeletons are often used to help in gait training of stroke patients. However, such robotic systems have so far yielded only mixed results in benefiting the clinical population. Therefore, there is a need to investigate how gait learning and de-learning get characterised in brain signals and thus determine neural substrate to focus attention on, possibly, through an appropriate brain-computer interface (BCI). To this end, this paper reports the analysis of EEG data acquired from six healthy individuals undergoing robot-assisted gait training of a new gait pattern. Time-domain partial Granger causality (PGC) method was applied to estimate directed neural connectivity among relevant brain regions. To validate the results, a power spectral density (PSD) analysis was also performed. Results showed a strong causal interaction between lateral motor cortical areas. A frontoparietal connection was found in all robot-assisted training sessions. Following training, a causal "top-down" cognitive control was evidenced, which may indicate plasticity in the connectivity in the respective brain regions.