Zero-Calibration MI Decoding via Self-Supervised Representation and Ensemble Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) is crucial for Brain-Computer Interface (BCI) research.
• Zero-calibration MI decoding simplifies BCI by removing the need for subject-specific models.
• Accurate extraction of invariant features is essential for effective zero-calibration MI decoding.

Purpose of the Study:

• To propose an efficient strategy for accurate extraction of intrinsic invariant features from EEG signals for zero-calibration MI decoding.
• To enhance model performance through a combination of self-supervised and supervised learning.
• To provide an innovative solution for cross-subject MI decoding.

Main Methods:

• Employs self-supervised representation learning via random masking and feature reconstruction to uncover universal EEG signal features.
• Utilizes an ensemble learning classifier for feature compression and enhanced model performance.
• Applies a pure convolutional neural network (CNN) architecture.

Main Results:

• Achieved accuracy exceeding 60% in a complex four-class MI classification task.
• Obtained accuracies of 85.50% and 82.98% in two binary classification tasks.
• Demonstrated statistically significant performance improvements compared to existing methods (p < 0.05).

Conclusions:

• The proposed method offers an innovative and efficient solution for cross-subject zero-calibration MI decoding.
• The combination of self-supervised learning and ensemble classification significantly enhances decoding accuracy.
• This approach reduces data annotation and training time, making BCIs more accessible.