EEG-Based Gait Phase Decoding From Combined Action Observation and Motor Imagery

Gait recovery is a crucial component of stroke rehabilitation. While Brain-Computer Interfaces (BCIs) decoding motor intent from motor imagery (MI) have shown success, their application in the area of gait phase decoding remains limited. Combining Action Observation (AO) and MI paradigms have demonstrated enhanced motor cortex activation compared to AO or MI alone. This study investigated the feasibility of decoding swing and stance phass of gait from electroencephalogaphy (EEG), via a proposed feature extraction and classification method. A novel dataset, utilizing the Combined AO, MI, and Steady-State Motion Visual Evoked Potential (SSMVEP) (CAMS-BCI) paradigm, was collected from twenty healthy volunteers. Employing an innovative labelling technique, three different classification methods were compared. Among them, broad band EEG features with a linear classifier achieved the highest average f1-score of 0.77 in gait phase classification. Additionally, the methods achieved an overall accuracy of 70% in classifying individual Swing and Stance phases based on the CAMS stimulus responses. These findings provide valuable insights for the development of novel BCI feedback mechanisms specifically targeting different phases of gait. Implementing them in future designs can potentially enhance gait recovery outcomes in post-stroke rehabilitation.