Tangent Space Features-Based Transfer Learning Classification Model for Two-Class Motor Imagery Brain-Computer...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) performance improves with more training data.
• Non-stationary data across sessions and subjects negatively impacts classifier generalization.
• Long calibration times are a significant barrier in BCI system deployment.

Purpose of the Study:

• To reduce the lengthy calibration period in BCI systems.
• To propose a transfer learning model for evaluating unseen single trials without prior training data.
• To enhance BCI usability and accessibility.

Main Methods:

• A novel transfer learning model combining a generalized multivariate empirical-mode decomposition (8-30 Hz band) preprocessing technique.
• Exploitation of tangent space features within the Riemannian geometry framework.
• Utilizing shared data structures across multiple sessions and subjects.

Main Results:

• The proposed model achieved comparable performance improvements across multiple subjects.
• Subject-specific calibration was not required, significantly reducing setup time.
• The method demonstrated effectiveness against state-of-the-art techniques.

Conclusions:

• The transfer learning approach effectively reduces BCI calibration time.
• The model generalizes well across subjects and sessions without compromising performance.
• This work paves the way for more efficient and user-friendly BCI systems.