Dynamic Eigenvector Centrality as a Biomarker for Motor Imagery Brain-Computer Interfaces

Brain functional connectivity relies on the evaluation of instantaneous similarity between different brain regions. This strategy has been widely applied in neuroscience using fMRI data in order to understand brain connectivity organization implicated in some of the main brain pathologies, such as Parkinson and Alzheimer’s disease. Recently, some studies have shown that the functional connectivity evaluation by means of graph metrics, more specifically eigenvector centrality, offers improved task discrimination in motor imagery EEG-based brain-computer interfaces. Nonetheless, these studies considered the connectivity as a static phenomenon and did not take into account the dynamic behaviour of the motor imagery process, which could add relevant information to task discrimination, by the inclusion of the preceding imagery (intention) and post-imagery phases. This work presents a classification performance comparison between dynamic eigenvalue centrality and dynamic power during the motor imagery experiment with a methodology based on a sliding window feature extraction with Pearson correlation as a measure of functional connectivity and a template matching classification approach, including preceding and post-motor imagery intervals. The results indicate that eigenvalue centrality can offer a promising complementary feature to classical bandpower for MI classification in BCI systems.