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

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 FMs 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.

Main findings

1. 1

   EEG-FMs show no consistent advantage under linear probing

   EEG-FMs performed similarly to supervised models and generic time-series foundation models (TS-FMs). The TS-FMs used neither EEG-specific architectures nor EEG data in pretraining.

2. 2

   Model rankings depend on the evaluation metric

   Model rankings based on a single metric differed from those based on other clinically grounded metrics. Higher scores on standard metrics did not necessarily correspond to better performance on clinically relevant measures.

3. 3

   Performance and rankings vary across datasets

   Both absolute performance and model rankings varied substantially across datasets within the same domain. Evaluation across multiple cohorts is therefore needed to assess how consistently a model performs.

BibTeX

@misc{kontras2026neuroatlas,
      title={NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces},
      author={Konstantinos Kontras and Trui Osselaer and Stylianos G. Mouslech and
              Angeliki-Ilektra Karaiskou and Guido Gagliardi and Thomas Strypsteen and
              Mohammad Hossein Badiei and Anku Rani and Maarten Vanmarcke and
              Miguel Bhagubai and Chanakya Ekbote and Jaedong Hwang and
              Christos Chatzichristos and Paul Pu Liang and Maarten De Vos},
      year={2026},
      eprint={2605.14698},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2605.14698},
}