EEG Channel Localization and Selection via Training with Noise Injection for BCI Applications.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) is vital for brain activity monitoring.
• High channel counts in EEG data lead to impractical usage and complex models.

Purpose of the Study:

• To address high dimensionality challenges in EEG data.
• To introduce an innovative EEG channel selection algorithm, LSvT-NI.
• To reduce channel count, model size, and computational complexity while preserving accuracy.

Main Methods:

• Developed LSvT-NI algorithm based on model training and noise injection.
• Validated the algorithm using EEGNet and the BCI Competition IV 2a dataset.
• Applied white noise and pink noise at 5dB SNR for channel selection.

Main Results:

• Achieved 77.3% and 72.7% channel reduction with white and pink noise, respectively.
• Reduced model size by 11.7% and 11% with white and pink noise.
• Decreased computational complexity by 86.9% and 71.8% with white and pink noise.

Conclusions:

• LSvT-NI effectively reduces EEG data dimensionality and complexity.
• The algorithm offers practical and cost-efficient solutions for EEG analysis.
• LSvT-NI maintains high classification accuracy, proving its utility in real-world applications.