Section: New Results

Optimal motor tasks for Sensorimotor BCI calibration

Participants: Fabien Lotte

SensoriMotor Rhythm (SMR)-based Brain-Computer Interfaces (BCI) are among the most used ElectroEncephaloGraphy (EEG) BCI systems. However, such systems have low performance and many of their users are “non-responders”. There is thus a need to understand the limitations of current SMR-BCI and to improve them. Many of them use machine learning. They are typically calibrated on EEG signals collected while the users are performing Motor Imagery (MI), i.e., imagining limb movements. Once calibrated, they also use MI as control strategy. However, for many first time users of SMR-BCI, performing MI is new and difficult, and they may be unable to perform clear MI. Thus, using MI for calibration may result in suboptimal EEG features and corresponding real-time feedback. Therefore, we aim at elucidating whether MI tasks are the best motor tasks to use for calibration and control in SMR-BCI. To do so, we collected EEG signals from subjects instructed to perform four different motor tasks and a rest task, for multiple trials. In particular, subjects have to 1) execute real feet movements; 2) imagine feet movements (walking); 3) observe feet movements (walking), in a first person view and 4) observe feet movements while imagining them at the same time. Preliminary results revealed that for some subjects, calibrating EEG spatial filters on real motor movements can lead to better performances with an MI-BCI than calibrating them on MI tasks. This thus warrant further investigation into the calibartion tasks in SMR-BCI. This preliminary work, in collaboration with RIKEN Brain Science Institute in Japan, was presented as a poster at RTFIN 2017 [48].