Brain-computer interfaces (BCIs) based on motor imagery electroencephalogram (MI-EEG) signals have been extensively applied in various neural rehabilitation scenarios. However, existing methods primarily focus on designing complex architectures to extract spatio-temporal features from MI-EEG signals, often neglecting the brain dynamics information embedded within them. This oversight leads to the extraction of redundant information, ultimately reducing decoding performance. To address these challenges, we design a multi-level topology-guidance network (ML-TGNet) that leverages topological brain synchronization information to more effectively extract features related to MI tasks. ML-TGNet specifically comprises a multi-level topology guidance module, a feature pool module, and a multi-branch decoding module. To evaluate its performance, extensive experiments are conducted on three publicly available MI datasets: the BCI Competition IV-2a dataset, the High Gamma dataset, and the OpenBMI dataset. ML-TGNet achieves classification accuracies of 82.33%, 96.42%, and 85.26% on these three datasets, respectively, outperforming current state-of-the-art models. These findings confirm the efficacy of using brain synchronization information to guide MI decoding, thereby opening a novel approach for EEG-based brain state decoding by integrating brain dynamics into deep learning.

Understanding and decoding human brain activity from electroencephalography (EEG) signals is a fundamental problem in neuroscience and artificial intelligence, with applications ranging from cognition and emotion recognition to clinical diagnosis and brain–computer interfaces. While recent EEG foundation models have made progress in generalized brain decoding by leveraging unified architectures and large-scale pretraining, they inherit a scale-agnostic dense modeling paradigm from NLP and vision. This design overlooks an intrinsic property of neural activity—cross-scale spatiotemporal structure. Different EEG task patterns span a broad range of temporal and spatial scales, from brief neural activations to slow-varying rhythms, and from localized cortical activations to large-scale distributed interactions. Ignoring this diversity may lead to suboptimal representations and weakened generalization ability. To address these limitations, we propose CSBrain, a Cross-scale Spatiotemporal Brain foundation model for generalized EEG decoding. CSBrain introduces two key components: (i) Cross-scale Spatiotemporal Tokenization (CST), which aggregates multi-scale features within localized temporal windows and anatomical brain regions into compact scale-aware token representations; and (ii) Structured Sparse Attention (SSA), which models cross-window and cross-region dependencies for diverse decoding tasks, further enriching scale diversities while eliminating the spurious dependencies. CST and SSA are alternately stacked to progressively integrate cross-scale spatiotemporal dependencies. Extensive experiments across 11 representative EEG tasks and 16 datasets demonstrate that CSBrain consistently outperforms both task-specific models and strong foundation baselines. These results establish cross-scale modeling as a key inductive bias for generalized EEG decoding and highlight CSBrain as a robust backbone for future brain–AI research.