A Subject-Independent Brain-Computer Interface Framework Based on Supervised Autoencoder.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery-based brain-computer interfaces (MI-BCIs) require user-specific calibration, which is time-consuming.
• Subject-dependent characteristics of MI signals pose challenges for developing universal BCI systems.
• Existing supervised learning methods for MI-BCI focus on discrimination, limiting the discovery of common signal patterns.

Purpose of the Study:

• To propose a subject-independent MI-BCI framework to eliminate the need for a calibration phase.
• To develop a novel supervised autoencoder (SAE) model for enhanced MI signal generalization.
• To improve the immediate usability of MI-BCI systems for naive users.

Main Methods:

• A supervised autoencoder (SAE) was developed for subject-independent MI-BCI.
• The proposed framework was validated using dataset 2a from BCI Competition IV.
• Performance was evaluated against conventional algorithms like Common Spatial Patterns (CSP) and Filter Bank Common Spatial Patterns (FBCSP).

Main Results:

• The proposed Subject-Independent Supervised Autoencoder (SISAE) model demonstrated superior performance.
• SISAE outperformed conventional BCI algorithms in mean Kappa value for eight out of nine subjects.
• The results indicate improved subject-to-subject generalization capabilities.

Conclusions:

• The SISAE model effectively circumvents the calibration phase in MI-BCI systems.
• This approach enhances the generalizability of MI-BCI across different users.
• The findings suggest a promising direction for developing more accessible and immediate-use BCI technologies.