Transferring Spatial Filters via Tangent Space Alignment in Motor Imagery BCIs

2025-04-23

AI Summary

To address poor cross-subject transferability in motor imagery brain–computer interfaces (MI-BCIs) and the weak generalization of conventional Common Spatial Pattern (CSP) spatial filters under limited training samples, this paper proposes a cross-subject CSP method based on Riemannian manifold tangent space alignment. For the first time, it integrates tangent space projection and alignment of covariance matrices on the Riemannian manifold into the CSP framework, constructing a subject-invariant discriminative filter space without relying on large amounts of subject-specific labeled data. By aligning and fusing covariance matrices across subjects on the manifold, the method achieves significantly higher classification accuracy than standard CSP under low-sample settings—and maintains robust improvements even with full datasets—on three public benchmarks. This approach breaks CSP’s strong dependence on subject-specific training data and establishes a generalizable geometric modeling paradigm for resource-constrained MI-BCIs.

Technology Category

• Machine Learning: Learning with Manifolds
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

We propose a method to improve subject transfer in motor imagery BCIs by aligning covariance matrices on a Riemannian manifold, followed by computing a new common spatial patterns (CSP) based spatial filter. We explore various ways to integrate information from multiple subjects and show improved performance compared to standard CSP. Across three datasets, our method shows marginal improvements over standard CSP; however, when training data are limited, the improvements become more significant.

Problem

• Improving subject transfer in motor imagery BCIs
• Aligning covariance matrices on Riemannian manifold
• Enhancing performance with limited training data

Innovation

• Align covariance matrices on Riemannian manifold
• Compute new CSP-based spatial filter
• Integrate information from multiple subjects