Addressing Performance and User Diversity Challenges in Motor Imagery-based Brain-computer Interfaces

Motor imagery-based brain–computer interfaces (MI-BCIs) show great promise but face
persistent challenges in real-world applications. Users often require lengthy calibration
sessions (20–30 minutes) to collect labeled data for training personalized models, and system
performance varies significantly across users and sessions. Notably, 10–30% of users are
unable to gain effective control, a phenomenon known as BCI inefficiency. Furthermore, the
underrepresentation of minority groups in BCI research exacerbates disparities in system
development and usability.
This thesis aims to enhance the usability and inclusivity of MI-BCIs through novel algorithmic
methods and community engagement. First, it introduces Scaled and Warped Common Spatial
Patterns (SW-CSP), a feature extraction technique that improves spatial filter estimation by
accounting for amplitude and temporal variability across trials. SW-CSP outperforms standard
CSP in classification accuracy.
The thesis also investigates pre-cue parietal alpha power as a neurophysiological marker of
user readiness and attention. This marker significantly correlates with motor imagery
performance across large datasets, offering a potential predictor for personalized training
strategies.
To address the issue of lengthy calibration, a new transfer learning framework, SW-TL-CSP, is
proposed. This method maps previously recorded data to new users via a computationally
efficient, non-linear alignment, followed by Fisher score-based feature selection. It achieves
strong performance even with as few as five to ten labeled trials per class, outperforming
existing methods in real-world scenarios.
Finally, the thesis engages the Black African community in Sheffield through workshops and
interviews to explore barriers to participation in neurotechnology. It demonstrates the
feasibility of conducting neuroimaging studies in community centers and highlights how
inclusive, community-based approaches can reduce participation gaps and broaden the
accessibility of BCI research.