Year: 2026
Type: preprint
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.
Source: arXiv (Cornell University)
Authors: Kontras, Konstantinos, Osselaer, Trui, Mouslech, Stylianos G., Karaiskou, Angeliki-Ilektra, Gagliardi, Guido, Strypsteen, Thomas, Badiei, Mohammad Hossein, Rani, Anku, Vanmarcke, Maarten, Bhagubai, Miguel, Ekbote, Chanakya, Hwang, Jaedong, Chatzichristos, Christos, Liang, Paul Pu, De Vos, Maarten
Institutions: Massachusetts Institute of Technology, KU Leuven
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
Funders: European High Performance Computing Joint Undertaking, National Science Foundation, American Sleep Medicine Foundation, Temple University, Vlaams Supercomputer Centrum, European Commission, Belgian Federal Science Policy Office, Fonds Wetenschappelijk Onderzoek, KU Leuven, China Scholarship Council, Innoviris, Vlaamse regering, National Institutes of Health, Defense Advanced Research Projects Agency, HORIZON EUROPE Framework Programme, National Heart, Lung, and Blood Institute, National Human Genome Research Institute, National Center for Advancing Translational Sciences, H2020 Marie Skłodowska-Curie Actions
Awards: G1261793039, G1276754516, G1294553593, Outcomes of Sleep Disorders in Older Men-Minneapolis, ASSOCIATION OF SLEEP DISORDERS WITH CARDIOVASCULAR HEALTH ACROSS ETHNIC GROUPS, Automatic discovery and processing of EEG cohorts from clinical records, The Neural Engineering Data Consortium: Building Community Resources to Advance Research, G1640378483, G1983464145, G2095942735, G2144938473, G22458031, G2726417417, Data-driven approaches for personalized medicine, G2837638605, Automated Analysis of Kidney Transplant Biopsies, G290884426, G291913881, Task- and device- agnostic Artificial Intelligence (AI) for EEG analysis, G3093995152, G3210485888, Familial Aggregation and Natural History of Sleep Apnea, G3409963932, G3581052147, G3595232572, G3667269547, G3990748671, Self-Supervised Representation Learning for analysis of wearable data from patients with epilepsy, G4309390892, G4401650913, Biometric Genetic Analysis of CardiopulmonaryDisease, G4590187187, Artificial Intelligence (AI) for data-driven personalised medicine, G4986157935, G5002562366, G5075046117, G5108002584, eXplainable Artificial Intelligence to embed new-generation Artificial Intelligence architectures for brain signal analysis in clinical scenarios, G5653954166, G5654179670, G5760880277, G5994613593, Data-driven approaches in computational cardiology, G66341962, G6742279884, G6856766826, Outcomes of Sleep Disorders in Older Men-Palo Alto, G7166352866, A Multicenter RCT on the impact of Wearable EEG and AI-Based Seizure Detection in absence seizure management, Outcomes of Sleep Disorders in Older Men, Advanced modelling of cerebrovascular autoregulation lab animal data to build a patient autoregulation monitor, G755179758, Outcomes of Sleep Disorders in Older Men, Outcomes of Sleep Disorders in Older Men-Birmingham, Methusalem-BioMedAI - Explainable and generative AI for accelerating biomedical discoveries: from genome to function, G78770644, STTR Phase I: Real-time Automatic Analysis of Electroencephalograms in an Intensive Care Environment Using Deep Learning, G805458286, G8145863542, Outcomes of Sleep Disorders in Older Men - Pittsburgh, G8280945325, G8298419571, G848352833, AI-based multimodal ambulatory seizure detection, G9000004549, G9000004784, G9000006353, G9000017360, G9000019818, G9000025517, G9000063950, G9000065698, G9000132197, G991334795
OpenAlex ID: W7161300711