Riemannian geometry metric-based visual feedback for BCI user training: towards exploratory learning of motor imagery

Abstract

Objective Despite their many emerging applications, practical use of brain–computer interfaces (BCIs) is often impeded by BCI-inefficiency, that is, the failure of the technology to decode neural modulations with sufficient accuracy. Recent evidence suggests that ineffective user training, namely feeding back to the user performance metrics that do not relate to the future performance of the BCI, may be obstructing users from learning how to produce machine-discernible sensorimotor rhythm modulations. Here, we use models of human skill acquisition to design a user-training interface to address these challenges.Approach We presented feedback via Riemannian geometry-based user performance metrics, which were validated via BCI simulation as bearing relation to future classifier performance. We subsequently evaluated the effect of the proposed feedback on users’ interpretation of their performance.Results Regression models showed that the metrics accounted for 53%–62% of intersubject variation in future classification accuracy with common BCI classifiers, thereby substantiating the use of the metric to guide user training. Participants were significantly better (p < 0.05) at detecting user performance changes with Riemannian metric-based feedback than with classifier feedback.Conclusion Our findings suggest that the proposed metrics can be effective for both assessing and communicating user performance, and therefore, warrant further investigation.

Keywords

• Brain–computer interface
• EEG
• sensorimotor rhythms
• motor imagery
• learning
• feedback interface