Data augmentation using masked principal component representation for deep learning-based SSVEP-BCIs.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Data augmentation improves deep learning for brain-computer interfaces (BCIs) using electroencephalography (EEG).
• Existing methods manipulate signals directly, risking distortion.
• Limited EEG data poses a challenge for robust model training.

Purpose of the Study:

• To introduce Masked Principal Component Representation (MPCR), a component-level data augmentation method for EEG-based BCIs.
• To address limitations of signal-level augmentation techniques.
• To enhance the robustness and classification accuracy of deep learning models.

Main Methods:

• MPCR employs a principal component-based reconstruction with random masking of principal components.
• Selected principal components are zeroed, introducing perturbations in reconstructed samples.
• Reconstruction via remaining components preserves inherent EEG signal structure, expanding training data diversity.

Main Results:

• MPCR significantly improves classification accuracy across various deep learning models.
• The method demonstrates superior performance compared to state-of-the-art augmentation techniques.
• MPCR exhibits high compatibility with existing BCI frameworks.

Conclusions:

• MPCR is a simple yet effective component-level data augmentation strategy.
• This technique offers valuable advancements for EEG-based BCI data augmentation.
• MPCR contributes to developing more robust deep learning models for BCIs.