Brain state differences between calibration and application periods influence BCI classification accuracy

Introduction: Non-invasive Brain-Computer Interfaces (BCI) translate a user's intent into technical control signals without relying on peripheral nerves and muscles. The Berlin-BCI application uses a calibration period to train machine learning algorithms for the classification of single-trial brain signals indicating specific motor intentions. These signals are then used in diverse application scenarios where the classifier output is fed back to the user, e.g., by a cursor movement. Here, we analyse possible reasons why the BCI classification accuracy can differ between calibration and feedback trials. Methods:…