Intra- and Inter-Subject Common Spatial Pattern for Reducing Calibration Effort in MI-Based BCI.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) require substantial labeled data for model calibration, limiting their practical use.
• Transfer learning (TL) shows promise for reducing data requirements, but standardized approaches are lacking.
• Motor imagery (MI) BCIs are a key application area affected by data scarcity.

Purpose of the Study:

• To develop a novel transfer learning (TL) algorithm for motor imagery (MI) brain-computer interfaces (BCIs).
• To enhance the robustness and reduce the data dependency of BCI classification models.
• To exploit both intra- and inter-subject similarities for improved feature signal extraction.

Main Methods:

• Proposed a Euclidean alignment (EA)-based intra- and inter-subject common spatial pattern (EA-IISCSP) algorithm to estimate four spatial filters.
• Developed a TL-based classification framework utilizing EA-IISCSP, Linear Discriminant Analysis (LDA), and Support Vector Machine (SVM).
• Evaluated the algorithm on two MI datasets, comparing its performance against three state-of-the-art TL algorithms.

Main Results:

• The EA-IISCSP algorithm significantly outperformed existing TL algorithms across various training data sizes (15-50 trials per class).
• The proposed method demonstrated a notable reduction in the required training data while maintaining high classification accuracy.
• The algorithm effectively enhanced the robustness of feature signals by leveraging subject similarities and variability.

Conclusions:

• The developed EA-IISCSP-based TL framework offers a significant advancement for MI-BCIs by reducing data calibration needs.
• This approach facilitates more practical and widespread application of BCI technology.
• The findings highlight the potential of exploiting subject-specific patterns for more efficient BCI systems.