LGL-BCI: A Motor-Imagery-Based Brain-Computer Interface with Geometric Learning

2023-10-12

arXiv.org

AI Summary

To address challenges in motor imagery brain–computer interfaces (BCIs)—including high amplitude/phase variability and complex spatial correlations in EEG signals, redundant model parameters, and excessive inference latency—this paper proposes the first lightweight geometric learning BCI framework tailored for Symmetric Positive Definite (SPD) manifolds. Methodologically, it introduces: (1) a differentiable, lossless dynamic channel selection module operating directly on the SPD manifold via spectral decomposition; and (2) a coupled architecture integrating lossless tensor transformations with a highly streamlined neural backbone to jointly preserve geometric discriminability and computational efficiency. Evaluated across two devices and two benchmark datasets, the framework achieves a mean classification accuracy of 82.54%—surpassing state-of-the-art methods by 20.32%—with only 64.9K parameters (a 64.7% reduction). This work is the first to empirically demonstrate the feasibility, robustness, and real-time decoding potential of geometric deep learning in resource-constrained BCI systems.

Abstract

Brain--computer interfaces are groundbreaking technology whereby brain signals are used to control external devices. Despite some advances in recent years, electroencephalogram (EEG)-based motor-imagery tasks face challenges, such as amplitude and phase variability and complex spatial correlations, with a need for smaller models and faster inference. In this study, we develop a prototype, called the Lightweight Geometric Learning Brain--Computer Interface (LGL-BCI), which uses our customized geometric deep learning architecture for swift model inference without sacrificing accuracy. LGL-BCI contains an EEG channel selection module via a feature decomposition algorithm to reduce the dimensionality of a symmetric positive definite matrix, providing adaptiveness among the continuously changing EEG signal. Meanwhile, a built-in lossless transformation helps boost the inference speed. The performance of our solution was evaluated using two real-world EEG devices and two public EEG datasets. LGL-BCI demonstrated significant improvements, achieving an accuracy of 82.54% compared to 62.22% for the state-of-the-art approach. Furthermore, LGL-BCI uses fewer parameters (64.9K vs. 183.7K), highlighting its computational efficiency. These findings underscore both the superior accuracy and computational efficiency of LGL-BCI, demonstrating the feasibility and robustness of geometric deep learning in motor-imagery brain--computer interface applications.

Problem

• Enhance EEG-based motor-imagery BCI accuracy
• Reduce model size and speed up inference
• Address EEG signal variability and complexity

Innovation

• Geometric deep learning architecture
• EEG channel selection module
• Lossless transformation for speed