System Filter-Based Common Components Modeling for Cross-Subject EEG Decoding

TL;DR

This work addresses the challenge of inter-subject variability in cross-subject MI-EEG decoding by introducing a system filter that expands a subject's brain system into a spectral relation spectrum, filters out personalized components, and reconstructs a purer system based on stable common components. Integrated into the CSD-SF framework, this approach uses a four-stage pipeline (APM, EPM, ECC, CCM) and a LOSO evaluation on the BCIC IV 2a dataset to derive a universal cross-subject model. The key contributions include formalizing the system filter on a polynomial relation spectrum, using a one-sample $t$-test to identify common components, and demonstrating a $3.28\%$ average accuracy improvement over baselines, along with ablation evidence of the filter's value. The results highlight improved robustness and generalizability for EEG-based BCI, with practical implications for reducing subject-specific calibration in real-world deployments.

Abstract

Brain-computer interface (BCI) technology enables direct communication between the brain and external devices through electroencephalography (EEG) signals. However, existing decoding models often mix common and personalized components, leading to interference from individual variability that limits cross-subject decoding performance. To address this issue, this paper proposes a system filter that extends the concept of signal filtering to the system level. The method expands a system into its spectral representation, selectively removes unnecessary components, and reconstructs the system from the retained target components, thereby achieving explicit system-level decomposition and filtering. We further integrate the system filter into a Cross-Subject Decoding framework based on the System Filter (CSD-SF) and evaluate it on the four-class motor imagery (MI) task of the BCIC IV 2a dataset. Personalized models are transformed into relation spectrums, and statistical testing across subjects is used to remove personalized components. The remaining stable relations, representing common components across subjects, are then used to construct a common model for cross-subject decoding. Experimental results show an average improvement of 3.28% in decoding accuracy over baseline methods, demonstrating that the proposed system filter effectively isolates stable common components and enhances model robustness and generalizability in cross-subject EEG decoding.

Figures (11)

• Figure 1: The research goals of this paper. (a) Previous decoding models mix common and personalized components, leading to poor generalization for new subjects. (b) The proposed approach filters out personalized components and builds a pure cross-subject model based only on common components, enabling stable decoding.
• Figure 2: Conceptual analogy between the proposed system filter and signal filtering. (a) Signal filtering. A complex signal with noise is mapped into the frequency domain via Fourier transform and decomposed into multiple components. Filtering out unnecessary frequencies (e.g., 120 Hz, 180 Hz) yields a clean waveform containing only the target components (e.g., 20 Hz, 50 Hz). (b) System filter. A complex system, represented as a sum of polynomial terms, can be transformed into a spectral representation. By removing unnecessary terms (e.g., $x_1$, $x_3$, $x_i^2$) and retaining only the target ones, a purer and more interpretable system is obtained.
• Figure 3: Overall model architecture of the proposed CSD-SF for cross-subject EEG decoding for MI. The proposed CSD-SF consists of four major modules. 1. APM. 2. ERS. 3. ECC. 4. CCM. m: the number of common relation items. O: the number of outputs. Q: query. K: key. V: value.
• Figure 4: The experimental paradigm for BCIC IV 2a
• Figure 5: Cross-subject experimental setting.