When should MI-BCI feature optimization include prior knowledge, and which one?

Motor imagery-based brain–computer interfaces (MI-BCIs) rely on interactions between humans and machines. The (learning) characteristics of both components are key to understand improve performances. Data-driven methods often used select/extract features with very little neurophysiological prior. Should they include prior knowledge which one? We study the relationship MI-BCI performances a commonly subject-specific Most Discriminant Frequency Band (MDFB). Our results showed correlation selected MDFB (mean width) While causal link could not be determined, online obtained using constrained algorithm --enforcing associated high performances– seemed higher than an unconstrained algorithm. Finally, we machine learning 1) predict from 2) select automatically optimal This way, for subjects either clearly distinct or no EEG patterns.