An EEG may contain sensitive information. Studies of EEG signals during rest and sleep show that they can reveal signs of sleep disorders, autism spectrum disorders, or alcoholism, for example. This creates a dilemma regarding the use of the data: While the data is of immense value in research—such as in the development of methods for the early detection of neurodegenerative diseases—it is simultaneously subject to the strictest data protection requirements. And for good reason, as research shows that individuals can be identified based on their unique “brainprints” in EEG datasets.

The “NEMO” project (Non-identifiability of Electroencephalograms and comparable sensor signals from medical care for Open Science), funded by the German Federal Ministry of Research, Technology, and Space (BMFTR), has spent the past three years addressing the need for EEGs to be available to science on the one hand, and ensuring adequate data protection on the other. The consortium investigated how personal data can be removed from the recordings while retaining information necessary for specific interpretation. In the project’s concrete application example, the classification of sleep phases and the identification of so-called sleep spindles in the EEG should remain possible even after anonymization.

The “NEMO” project was coordinated by the Mobile Neurotechnologies group at the Fraunhofer Institute for Digital Media Technology IDMT in Oldenburg. The group has been researching and developing mobile EEG systems for many years and, with its expertise in the field of sleep monitoring, provided the specific application scenarios for the project. Dr.-Ing. Insa Wolf, Head of Mobile Neurotechnologies, highlights the project’s high relevance: “Mobile EEG technologies enable measurements outside of laboratories and clinics. Such systems are already finding their way into the consumer market, for example for sleep monitoring. Given the wealth of information provided by EEGs, data protection risks must be weighed against the opportunities for gaining insights in the healthcare sector. We should make full use of what is technically possible in terms of data protection.”

AI Technologies for secure EEG data

At the Fraunhofer IDMT headquarters in Ilmenau, experts in technical data protection designed methods for re-identifying individuals within large EEG datasets. They developed a machine learning algorithm that learns personal patterns within the complex biosignals and recognizes them in other datasets with an accuracy rate of over 80 percent. In reality, therefore, it becomes particularly critical when EEG data is published alongside the names of the participants: In such cases, measurements can be unambiguously assigned to a specific person and potentially linked to other datasets.

“The result underscores the need for special protective measures for EEG data. We therefore worked on AI algorithms in the project that modify EEG recordings so that they no longer allow any conclusions to be drawn about the participants, while still retaining their usefulness for the selected application areas,” explains Ilmenau-based data protection expert Thomas Köllmer. In the project, the signals were transformed to such an extent that only the characteristics necessary for automatic screening of sleep phases remained recognizable.

Christian-Albrechts-Universität zu Kiel and the University Medical Center Schleswig-Holstein Campus Kiel provided the necessary sleep data for the project. They also contributed their expertise in interpreting medical data and in data exchange within the framework of the data integration centers of the Medical Informatics Initiative.

Significant benefits for science and society

To examine the research approaches in a context as close to real-world applications as possible, stakeholder interviews and workshops were integrated into various phases of the project. The consortium collaborated with KIZMO GmbH (Clinical Innovation Center for Medical Technology Oldenburg) on a subcontract basis. In addition, expert reviews of the technological demonstrator were conducted. The demonstrator clearly illustrates how EEG data is anonymized and what implications this has for specific application scenarios. The platform was developed as a proof-of-concept in collaboration with Ascora GmbH from Ganderkesee.

With the completion of the “NEMO” project in December 2025, the consortium has demonstrated the feasibility of anonymization technologies for securing biosignals. The experts consider further research and development of such technologies to be highly relevant. “The further development of these anonymization approaches is important now because it strengthens trust in medical and non-medical applications and paves the way for broader data availability in the spirit of open science in research and teaching,” summarizes data protection expert Thomas Köllmer.

Publications related to the project:

J. Scanlon, A. Pelzer, M. Gharleghi, K. Fuhrmeister, T. Köllmer, P. Aichroth, R. Göder, C. Hansen, I. Wolf, 2025:“What your brain activity says about you: A review of neuropsychiatric disorders identified in resting-state and sleep EEG data”, arXiv preprint arXiv:2510.04984.

M. Gharleghi, K. Fuhrmeister, T. Köllmer, A. Pelzer, J. Scanlon, I. Wolf, J. Lechinger, R. Göder, 2025: “What Your Brain Activity Says About You: EEG-based Subject Verification Across Sessions”, in 2025 10th International Conference on Frontiers of Signal Processing (ICFSP) (pp. 55-59). IEEE.

K. Fuhrmeister, A. Pelzer, F. Radke, J. Lechinger, M. Gharleghi, T. Köllmer, I. Wolf, 2025: “Bridging privacy and utility: Synthesizing anonymized EEG with constraining utility functions”, Accepted for EUSIPCO 2026, preprint available at arXiv:2509.20454