When Brain Foundation Model Meets Cauchy-Schwarz Divergence: A New Framework for Cross-Subject Motor Imagery Decoding

Abstract

Decoding motor imagery (MI) electroencephalogram (EEG) signals, a key non-invasive brain-computer interface (BCI) paradigm for controlling external systems, has been significantly advanced by deep learning. However, cross-subject MI-EEG decoding remains challenging due to substantial inter-subject variability and limited labeled target data, which necessitate costly calibration for new users. Many existing multi-source domain adaptation (MSDA) methods indiscriminately incorporate all available source domains, disregarding the large inter-subject differences in EEG signals, which leads to negative transfer and excessive computational costs. Moreover, while many approaches focus on feature distribution alignment, they often neglect the explicit dependence between features and decision-level outputs, limiting their ability to preserve discriminative structures. To address these gaps, we propose a novel MSDA framework that leverages a pretrained large Brain Foundation Model (BFM) for dynamic and informed source subject selection, ensuring only relevant sources contribute to adaptation. Furthermore, we employ Cauchy-Schwarz (CS) and Conditional CS (CCS) divergences to jointly perform feature-level and decision-level alignment, enhancing domain invariance while maintaining class discriminability. Extensive evaluations on two benchmark MI-EEG datasets demonstrate that our framework achieves average accuracies of 86.17% and 78.41%, outperforming a broad range of state-of-the-art baselines. Additional experiments with a large source pool validate the scalability and efficiency of BFM-guided selection.

Figures (7)

• Figure 1: Overview of the proposed BFM-MSDA framework for cross-subject motor imagery decoding. The framework comprises: (a) a source selection phase, where LaBraM, a representative brain foundation model, is employed to extract and hierarchically aggregate generic feature representations $\mathbf{h}_1, \mathbf{h}_2, \ldots, \mathbf{h}_S$ from $S$ source subjects and $\mathbf{h}_T$ from the target subject. Pairwise CS divergences $d_{S_i,T}$ are computed between each source $S_i$ and the target, and only sources with $d_{S_i,T}$ smaller than a predefined threshold $\delta$ are selected for adaptation; (b) a multi-source domain adaptation (MSDA) phase, in which feature-level alignment (FLA) is achieved by minimizing weighted CS divergences of the feature distributions $p(z)$ between each selected source and the target, as well as between all pairs of selected sources. Decision-level alignment (DLA) is enforced by minimizing conditional CS divergences between the conditional label distributions $p(y|z)$ of each source-target pair and all source-source pairs. (c) the testing phase, where the trained feature extractor and classifier infer MI class labels, such as left-hand and right-hand movement, for the target subject.
• Figure 2: Effect of selection threshold on performance and training time.
• Figure 3: Cross-subject classification accuracies of ablation variants on both datasets. (a) Dataset I. (b) Dataset II.
• Figure 4: Heat maps and MDS visualizations of pairwise distances for Datasets I and II. (a) Heat map of inter-subject distances for Dataset I. (b) MDS visualization of inter-subject distances for Dataset I. (c) Heat map of inter-subject distances for Dataset II. (d) MDS visualization of inter-subject distances for Dataset II.
• Figure 5: t-SNE visualization of target domain features extracted by the feature extractor for subject 1 from Dataset I (top row) and Dataset II (bottom row). (a) and (d) baseline EEGNet without DA, (b) and (e) MSDA without source selection, and (c) and (f) full BFM-MSDA framework.