braindecode/epint-pretrained

EpiNT — Epilepsy Neurophysiological Transformer

Pretrained weights for braindecode.models.EpiNT, a braindecode port of EpiNT (Zhang et al., 2025), a channel-independent transformer pretrained on 2,700 hours of scalp EEG and intracranial EEG (ECoG, sEEG) from 1,199 patients.

• Paper: Cross-Modal Epileptic Signal Harmonization: Frequency Domain Mapping Quantization for Pre-training a Unified Neurophysiological Transformer (arXiv:2506.17068)
• Original code & weights: RunKZhang/EpiNT
• braindecode docs: https://braindecode.org

Provenance & license

These weights are the authors' pretrained backbone, weights/representations.bin at commit 781e1d61 (sha256 b4449974058fd6d314debeef0a09c05730a70e822f17244e88d1ee4e7d9a8b1d), converted into the braindecode EpiNT module through its mapping (all 102 backbone tensors; the pretraining mask token embed.mask_encoding is dropped). The classification head is not part of the release and is randomly initialized.

The original release is under the MIT license, which this repository preserves:

MIT License

Copyright (c) 2024 Auton Lab, Carnegie Mellon University

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Model configuration

This checkpoint uses the paper's configuration (~19.2M parameters):

patch_size	256
embed_dim	512
ffn_dim	2048
n_layers	6
n_heads	8
n_times	3072 (12 patches)

Each encoder LayerNorm spans tokens and features, so keep n_times=3072: 12 s at 256 Hz for scalp EEG, or 3 s at 1024 Hz for iEEG. Signals are expected in microvolts. Every channel is encoded alone and the class tokens are averaged, so n_chans is free; n_outputs resets the task head.

Training data

The data that shaped these weights. The ids are the datasets: field of this card, which braindecode reads into model.card_data.datasets when you call from_pretrained (MOABB class names for MOABB datasets, OpenNeuro / NEMAR accessions, Hub ids otherwise). nm/on ids are datasets on NEMAR (on = NEMAR's copy of the OpenNeuro dataset with the same number). Rows marked private: are data the authors did not release; they are not in datasets:.

datasets: id	Dataset	Stage	Modality
on003029	OpenNeuro ds003029 (Li et al., 2021, 'Neural fragility' epilepsy iEEG)	pretraining	iEEG
on003498	OpenNeuro ds003498 (Fedele et al., 2017, HFO intraoperative ECoG)	pretraining	iEEG
on003555	OpenNeuro ds003555 (Cserpan et al., 2021, pediatric epilepsy scalp EEG 10-20)	pretraining	EEG
on003844	OpenNeuro ds003844 (Demuru et al., 2022, RESPect intraoperative/long-term iEEG)	pretraining	iEEG
on003876	OpenNeuro ds003876 (Gunnarsdottir et al., 2023, Epilepsy iEEG interictal multicenter dataset)	pretraining	iEEG
on004100	OpenNeuro ds004100 (Bernabei et al., 2022, normative iEEG maps, HUP)	pretraining	iEEG
on004752	OpenNeuro ds004752 (Dimakopoulos et al., 2022, verbal working memory iEEG/EEG)	pretraining	iEEG
on005398	OpenNeuro ds005398 (Zhang et al., 2022, UCLA HFO iEEG)	pretraining	iEEG
TUEP	TUH EEG Epilepsy Corpus	pretraining	EEG
TUSZ	TUH EEG Seizure Corpus	pretraining	EEG
nm000385	Siena Scalp EEG Database (Detti et al., 2020)	pretraining	EEG
ZJNU	JHCH and JHMCHH epileptic EEG dataset (Zhejiang Normal University; Chen et al., 2025), named ZJNU in the EpiNT paper	pretraining	EEG
AUB	Epileptic EEG dataset (Nasreddine, 2021, Mendeley Data 5pc2j46cbc; American University of Beirut), named AUB in the EpiNT paper	pretraining	EEG

• 13 public sources, 2,741.1 h, 1,199 patients (2,548.3 h scalp EEG, 165.5 h iEEG); TUEP restricted to patients with confirmed epilepsy
• Sources: arXiv:2506.17068v1 Section 2.1; arXiv:2506.17068v1 Table 1 'Summary of pre-training datasets' (p.20), row 1; arXiv:2506.17068v1 Table 1 (p.20), row 8; https://github.com/RunKZhang/EpiNT/blob/781e1d61e72899d59b4955b4c075404eef532a40/configs/dataset_configs.yaml.

Usage

from braindecode.models import EpiNT

model = EpiNT.from_pretrained("braindecode/epint-pretrained", n_chans=17, n_outputs=2)

Citation

@article{zhang2025epint,
  title   = {Cross-Modal Epileptic Signal Harmonization: Frequency Domain Mapping Quantization for Pre-training a Unified Neurophysiological Transformer},
  author  = {Zhang, Runkai and Yu, Hua and Gan, John Q. and Wang, Haixian},
  journal = {arXiv preprint arXiv:2506.17068},
  year    = {2025}
}

Cite Braindecode

If you use this braindecode port in scientific work, cite the software on Zenodo, doi:10.5281/zenodo.17699192, and the original model and datasets cited above.

@software{braindecode,
  author = {Aristimunha, Bruno and
            Guetschel, Pierre and
            Wimpff, Martin and
            Gemein, Lukas and
            Rommel, Cedric and
            Banville, Hubert and
            Sliwowski, Maciej and
            Wilson, Daniel and
            Brandt, Simon and
            Gnassounou, Théo and
            Paillard, Joseph and
            {Junqueira Lopes}, Bruna and
            Sedlar, Sara and
            Moreau, Thomas and
            Chevallier, Sylvain and
            Gramfort, Alexandre and
            Schirrmeister, Robin Tibor},
  title = {Braindecode: toolbox for decoding raw electrophysiological brain data
           with deep learning models},
  url = {https://github.com/braindecode/braindecode},
  doi = {10.5281/zenodo.17699192},
  publisher = {Zenodo},
  license = {BSD-3-Clause},
}