Adaptation in motor imagery brain-computer interfaces and its implication in rehabilitation

In BCI based stroke rehabilitation systems, motor imagery is detected and fed back to patients via various modalities, being visual, robotic, haptic, and so on. Higher rate of detecting motor imagery is desired for potentially better rehabilitation outcome, as it affects training intensity, patients' engagement and contingent feedback effect. It is a long-standing challenge in BCI research as how to develop a robust BCI system, which can produce highly accurate detection results across multiple sessions despite large variations and interferences in real-world environments. Although some of the factors causing performance to vary can be minimized or mitigated, by for example using wireless portable brain signal recoding equipment, restraining body movements etc, many other factors, especially endogenous ones, are unavoidable. Therefore, methods which are able to keep tracking variations in brain signal and adapting models for motor imagery detection are highly useful. In this talk, we will discuss various adaption schemes and their implications in stroke rehabilitation using EEG based braincomputer interface.