Increasing session-to-session transfer in a brain-computer interface with on-site background noise acquisition

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) typically require extensive training sessions to create classifiers.
• Dynamic variations in brain signals and noise necessitate frequent retraining, limiting BCI practicality.

Purpose of the Study:

• To investigate if background noise characteristics can be leveraged for session-to-session data transfer in BCIs.
• To reduce or eliminate the need for repeated training phases in BCI systems.

Main Methods:

• Collected background noise data during brief, on-site sessions to characterize noise dynamics.
• Implemented a session-to-session transfer strategy using a regularized spatiotemporal filter (RSTF).
• Evaluated the RSTF with and without bias correction (BC) on 20 cross-session datasets from 12 subjects.

Main Results:

• Proposed session-to-session strategies achieved performance comparable or slightly lower than conventional methods requiring per-session training.
• The RSTF, particularly with bias correction, demonstrated superior performance in session-to-session transfers compared to existing approaches.

Conclusions:

• Session-to-session transfer is feasible for BCIs by incorporating on-site background noise suppression.
• This approach potentially eliminates the need for extensive retraining, enhancing BCI system feasibility.