Harnessing Spectrum Video for Subject-Level Few-Shot and Cross-Montage EEG Generalization

Existing EEG models are limited by electrode heterogeneity and rigid "channel-first" architectures that treat sensors as independent features. We propose Brain Signal Rendering (BSR), which reinterprets EEG as a physical projection of neural activity and transforms raw signals into structured spatiotemporal tensors (termed Spectrum Videos), enabling the transfer of rich priors from video foundation models. By utilizing VideoMAE for self-supervised pre-training, BSR learns robust, layout-agnostic spatiotemporal representations that preserve neural topology. We further employ subject-level few-shot learning and introduce cross-montage fine-tuning to rigorously evaluate generalization across subjects and electrode configurations. Experiments show that VideoMAE model integrated with the BSR framework significantly outperforms state-of-the-art spectrum based methods, providing a scalable and data-efficient foundation for generalizable EEG modeling. Our code is available at https://github.com/yanweifu-sii/BSR-VideoMAE.

Electroencephalography (EEG) is a powerful tool for measuring brain activity, but it faces a major challenge: different devices use different numbers and arrangements of sensors (electrodes). This makes it hard to train a single model that works reliably across hospitals, labs, or even different people. Our paper introduces a new way to think about EEG signals—treating them like a video of brain activity rather than a disconnected set of sensor readings. We convert EEG data into “spectrum videos” that preserve both the timing of brain signals and the physical layout of electrodes on the scalp. This allows us to leverage existing video artificial intelligence models, which are already very good at learning patterns from moving images. We show that this approach helps AI systems adapt to new individuals using only a few seconds of data and work effectively even when the electrode setup changes completely.