NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces

2026-05-14

AI Summary

This study addresses the lack of a unified, clinically relevant evaluation benchmark for electroencephalography (EEG) foundation models by introducing NeuroAtlas—the largest EEG benchmark to date, encompassing 42 datasets and over 260,000 hours of recordings across four key clinical tasks: epilepsy detection, sleep staging, brain age estimation, and brain–computer interfacing. The benchmark incorporates task-specific clinical metrics, including event-level decision criteria, hypnogram-derived features, and brain age gap. Through systematic comparison of EEG-specific and general-purpose time-series foundation models, the study finds that current EEG-tailored models do not significantly outperform generic ones, model performance varies substantially across tasks, and conventional machine learning metrics often fail to capture clinical utility. These findings highlight the ongoing challenges in achieving generalizable EEG representations and establish a clinically grounded evaluation framework for future model development.

Abstract

Foundation models (FMs) promise to extract unified representations that generalize across downstream tasks. They have emerged across fields, including electroencephalography (EEG), but it is less clear how effective they are in this particular field. Published evaluations differ in datasets, in the EEG-specific preprocessing that might influence reported results, and in the reported metrics, frequently obscuring the clinical relevance in EEG. We introduce NeuroAtlas, the largest EEG benchmark to date: 42 datasets and 260k hours covering clinical EEG (epilepsy, sleep medicine, brain age estimation) and brain-computer interfaces, and include multiple datasets per task along with bespoke clinical evaluation metrics. Besides evaluating EEG-FMs with respect to supervised baselines, we present results from generic time-series FMs. We report three findings. First, EEG-specific FMs do not consistently outperform time-series FMs, which have neither EEG-focused architectures nor been pretrained on EEG. Second, standard machine learning metrics are insufficient to assess clinical utility: thus, we thoroughly evaluate more appropriate measures such as the quality of event-level decision-making, hypnogram-derived features, and the brain-age gap in the domains of epilepsy, sleep, and brain age, respectively. Third, model rankings and performance can vary substantially within domains. We conclude that pretrained models perform largely on par, with only narrow advantages for a few, and that current models do not yet deliver on the promise of an out-of-the-box unified EEG model. NeuroAtlas exposes this gap and provides the datasets and metrics for the next generation of unified EEG FMs.

Problem

• foundation models
• clinical EEG
• brain-computer interfaces
• benchmarking
• clinical utility

Innovation

• Foundation Models
• EEG Benchmarking
• Clinical Evaluation Metrics
• Time-series Modeling
• Brain-Computer Interfaces