MI-CES: An explainable weak labelling approach to example selection for Motor Imagery BCI classification.

Area of Science:

• Neuroscience
• Computer Science
• Rehabilitation Engineering

Background:

• Brain Computer Interfaces (BCI) enable control of assistive devices like wheelchairs.
• BCI systems require user-machine training for adaptation and control accuracy.
• User feedback during training can enhance BCI performance, but incorrect feedback may hinder adaptation.

Purpose of the Study:

• To propose MI-CES, an explainable example selection approach for motor imagery (MI).
• To address the issue of incorrect feedback negatively impacting user adaptation in BCI systems.
• To improve classification accuracy in MI-based BCI through intelligent example selection.

Main Methods:

• Development of MI-CES, an explainable example selection method based on motor imagery principles.
• Utilized two distinct classification techniques to evaluate the proposed method.
• Tested the approach on three diverse datasets, including multi-participant and multi-session recordings.

Main Results:

• Achieved a statistically significant increase in classification accuracy.
• Demonstrated the effectiveness of MI-CES across multiple datasets and participant variations.
• Validated the approach's ability to enhance BCI performance compared to standard feedback methods.

Conclusions:

• MI-CES offers a novel and explainable strategy for optimizing motor imagery BCI training.
• The proposed example selection method leads to improved classification accuracy and potentially better assistive device control.
• This approach mitigates negative adaptation effects from incorrect feedback, enhancing BCI system reliability.