EEG foundation models are usually pretrained with a fixed channel vocabulary or a limited set of montages, which makes transfer difficult when electrode layouts change. To address this, we introduce CortexBridge, a lightweight adapter that combines EEG features with electrode and atlas coordinates to map arbitrary montages into a shared cortical latent space.
CortexBridge was evaluated on three frozen foundation models—EEGPT, LaBraM, and CBraMod—using five BCI datasets from the Mother of All BCI Benchmarks (MOABB). The method improved performance in 13 out of 15 evaluations. Average balanced‑accuracy gains were 0.80% for EEGPT, 0.70% for LaBraM, and 3.26% for CBraMod, with a maximum gain of 13.02% on a 12‑class steady‑state visual evoked potential (SSVEP) classification task.
Visualization of the learned atlas representations revealed task‑dependent spatial patterns: SSVEP showed a concentrated representation in the Yeo Visual network, whereas auditory P300 was more diffusely distributed.
These findings establish cortical alignment as a learnable, anatomically grounded routing mechanism that can bridge heterogeneous EEG montages to pretrained foundation models.
Review