Multimodal Pretraining for Generalizable EEG Representation Learning

2026-07-23

AI Summary

This study addresses the limited generalization of existing EEG-based epilepsy detection models across datasets and subjects. To overcome this challenge, the authors propose the first multimodal foundation model that jointly leverages raw EEG signals, time-frequency images, and textual metadata, learning transferable representations through self-supervised pretraining in a shared embedding space. The method innovatively integrates a Mamba architecture for temporal signal encoding with a Vision Transformer (ViT) for time-frequency image processing, augmented by cross-view contrastive alignment and temporal consistency losses to enable label-free, generalizable representation learning. Experimental results demonstrate state-of-the-art performance on the CHB-MIT dataset, achieving an AUROC of 0.874 with a single model and 0.878 with ensembling. In leave-one-subject-out evaluation, the model attains an average balanced accuracy of 0.558, significantly improving cross-subject generalization.

Abstract

Electroencephalography (EEG) models used for epilepsy are often limited to specific datasets and tasks. This limited approach can make it challenging to apply these models across different datasets or in various situations. However, recent studies in foundation models and self-supervised learning suggest that an adaptable EEG backbone could support a range of EEG related tasks. In this study, we have developed a multimodal EEG foundation model that combines a raw signal encoder based on the Mamba architecture, a Vision Transformer (ViT)-style encoder for time-frequency data, and a lightweight encoder for text, all within a shared embedding space. The pretraining process relies on several innovative techniques, such as masked modeling, cross-view contrastive alignment, and temporal consistency losses. These methods are designed to create rich, seizure-relevant representations without requiring labeled data. To assess the efficacy and generalization of our pretrained model, we fine-tuned it on the canonical CHB-MIT seizure detection benchmark and additional seizure detection datasets, and conducted extensive experiments comparing different model variants. On the standard CHB-MIT split, our best single model achieved an AUROC of 0.874, and an ensemble variant reached 0.878 AUROC, representing state-of-the-art performance on this benchmark. In addition to standard train-test splits, we evaluated performance under a leave-one-subject-out (LOSO) protocol, which is rarely reported in prior EEG seizure modeling work and highlights the difficulty of patient-independent seizure detection, with a mean LOSO balanced accuracy of 0.558 across 19 subjects. Across datasets and evaluation settings, our multimodal foundation model enabled robust seizure detection and straightforward adaptation to new seizure detection scenarios, while also supporting interpretable seizure localization.

Problem

• EEG
• seizure detection
• generalization
• multimodal learning
• foundation model

Innovation

• multimodal pretraining
• Mamba architecture
• self-supervised learning
• cross-view contrastive alignment
• EEG foundation model