license: cc-by-nc-4.0
MegNIST
MegNIST is a magnetoencephalography (MEG) dataset for benchmarking non-invasive inner-speech decoding. The dataset contains recordings from a single participant performing repeated trials involving the digits 0–9.
The raw MEG data are organised according to the Brain Imaging Data Structure (BIDS). Processed data suitable for machine-learning analyses are provided separately in the derivatives/serialised/ directory.
Licence
MegNIST is licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) licence.
It is freely available for research and other non-commercial uses, subject to the terms of the licence. When sharing or redistributing MegNIST, including modified or derived versions, please credit the MegNIST authors, link to the dataset, and identify the CC BY-NC 4.0 licence. If you modify the dataset, please indicate that changes were made. For commercial licensing enquiries, please contact Oxford University Innovation (OUI).
Citation
If you use MegNIST, please cite:
Kwon, T., Somaiya, P., Elvers, G., Mantegna, F., Camera, A., Braeutigam, S., Woolrich, M. and Parker Jones, O. (2026). MegNIST: A Benchmark for Non-Invasive Inner Speech Decoding.
@misc{kwon2026megnist,
title={{MegNIST}: A Benchmark for Non-Invasive Inner Speech Decoding},
author={Kwon, Teyun and Somaiya, Pratik and Elvers, Gereon and Mantegna, Francesco and Camera, Anna and Braeutigam, Sven and Woolrich, Mark and Parker Jones, Oiwi},
year={2026}
}