How low can you go: evaluating electrode reduction methods for EEG-based speech imagery BCIs.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Speech imagery brain-computer interfaces (SI-BCIs) decode imagined speech from brain activity using non-invasive methods like electroencephalography (EEG).
• Current SI-BCIs often require high-resolution EEG systems (64+ electrodes), limiting practical application due to setup complexity and cost.

Purpose of the Study:

• To evaluate electrode reduction algorithms for EEG-based SI-BCIs.
• To identify optimal electrode configurations for efficient and accurate speech imagery decoding.
• To assess the feasibility of reducing system complexity for real-world SI-BCI deployment.

Main Methods:

• Applied various electrode reduction algorithms to three distinct EEG-based speech imagery datasets.
• Investigated different feature extraction and classification techniques in conjunction with reduced electrode setups.
• Analyzed classification accuracy and electrode distribution patterns across datasets.

Main Results:

• A 50% reduction in EEG channels was achievable without significant loss of classification accuracy across all tested datasets.
• Relevant brain activity for speech imagery was found to be distributed across the cortex, not solely in the left hemisphere.
• No consistent optimal electrode set was identified, suggesting high inter-subject variability.

Conclusions:

• Reduced-electrode SI-BCI systems are viable, offering a practical alternative to high-resolution setups.
• Individual tailoring of electrode configurations is necessary for optimal SI-BCI performance.
• Findings support the development of more compact and user-friendly SI-BCIs for broader adoption.