Self-correcting brain computer interface based on classification of multiple error-related potentials

<i>Objective.</i>Electroencephalogram (EEG) based brain-computer interfaces (BCIs) have shown tremendous promise in facilitating direct non-invasive brain-control over external devices. However, their practical application is hampered due to errors in command interpretation. A promising strategy for improving BCI accuracy is based on detecting error-related potentials (ErrPs), which are EEG potentials evoked in response to errors. Thus, performance can be improved by undoing actions that evoke potentials that the BCI detects as ErrPs. To achieve further improvement, we aimed to classify the type of error and correct, rather than just undo, erroneous actions. The objectives of this study are to develop an error classifier (EC) and to investigate the hypothesis that correcting the actions according to the EC decisions improves performance.<i>Approach.</i>To evaluate our hypothesis we developed a BCI application to control the pose of virtual hands with three possible commands: change the pose of either the right or left hand and maintain pose. Thus, when an action elicits an ErrP, the identity of the correct command is still undecided. The self-correcting BCI included an EC and was developed in three phases: hand control, initial brain control and self-correcting brain control. The first two phases were conducted by 22 participants, and half of them (<i>n</i>= 11) also completed the last phase.<i>Main results.</i>Detecting the type of error and correcting actions accordingly improved the success rate of the self-correcting BCI for each participant (<i>n</i>= 11), with a significant average improvement of 6.6%and best improvement of 13.5%.<i>Significance.</i>Self-correction, based on an EC, was demonstrated to improve the accuracy of BCIs for three commands. Thus, our work presents a significant step toward the development of more reliable and user-friendly non-invasive BCIs.