A brain-computer interface is a type of interface that translates users brain activities into commands for computers or machines. Among different brain activities, electroencephalography (EEG, i.e., the electrical waves of human brains) is the most widely used due to its relatively low cost, easiness to set up and portability. However, there is a large variety in EEG signals across users, or even across sessions within the same individual. Therefore, to achieve a robust performance, a BCI system usually requires a user to go through a long calibration process before each use to fine tune the statistical model used in the system. This calibration process severely hinders the practicality of BCI systems.
In this project we propose a closed-loop BCI framework that can detect errors made by the BCI classifier by measuring user’s brain responses to unexpected outcomes, so called Error-related Potential (ErrP). This error signal can be used as feedback signals to the BCI classifier for adaptively tuning it. Results suggest that with the proposed closed-loop BCI framework, a user can use a BCI system starting from a statistical model trained with data from other individuals, or previous sessions, and does not need to through the tiring calibration process. The model can be fine-tuned for this user along with the system being used.