Recruiting neural field theory for data augmentation in a motor imagery brain-computer interface.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) are crucial for restoring function but are limited by insufficient training data, impacting accuracy.
• Motor imagery tasks are a common BCI paradigm, but data scarcity hinders robust model training.

Purpose of the Study:

• To introduce and evaluate a novel data augmentation technique for BCI training data using neural field theory (NFT).
• To generate artificial EEG time series based on a corticothalamic NFT model to supplement limited real-world data.

Main Methods:

• Applied a corticothalamic NFT model to EEG data from the BCI Competition IV '2a' dataset for motor imagery tasks.
• Fitted NFT model parameters to common spatial patterns of motor imagery classes and generated synthetic EEG time series for augmentation.
• Evaluated the impact of NFT-based data augmentation on classification accuracy using specific signal features.

Main Results:

• Achieved significant accuracy improvements exceeding 2% for classifying the 'total power' feature using NFT-augmented data.
• Observed no significant accuracy improvement for the 'Higuchi fractal dimension' feature, suggesting model-feature specificity.
• Demonstrated the potential of biophysically accurate artificial data generated via NFT for enhancing BCI performance.

Conclusions:

• NFT-based data augmentation offers a promising approach to overcome data limitations in BCIs, particularly for certain signal features.
• The study highlights the importance of considering feature-model interactions when applying biophysically informed generative models.
• Further research into NFT applications can lead to more robust and accurate BCIs.