HUP iEEG Epilepsy Dataset

Description

The HUP iEEG Epilepsy Dataset is a de-identified intracranial-EEG corpus collected during the surgical workup of drug-resistant focal epilepsy at the Hospital of the University of Pennsylvania (HUP), released on OpenNeuro as ds004100 (Bernabei, Li, Revell et al., 2022). It comprises recordings from 58 adult patients who underwent intracranial monitoring with either subdural grids/strips plus depth electrodes (ECoG) or purely stereotactically placed depth electrodes (SEEG), and who subsequently received surgical resection or laser ablation. Each subject’s data carries both ictal (seizure) and interictal (non-seizure) recordings, electrode localizations in MNI space, clinically determined seizure-onset channels, and resection/ablation-zone annotations — the dataset was assembled specifically to support quantitative, network-based approaches to predicting epilepsy-surgery outcomes (Bernabei et al., Brain 2023).

Citation

• Paper DOI: 10.1093/brain/awad007
• Dataset DOI: 10.18112/openneuro.ds004100.v1.1.3
• Reference: Bernabei JM, Li A, Revell AY, Smith RJ, Gunnarsdottir KM, Ong IZ, Davis KA, Sinha N, Sarma S, Litt B. HUP iEEG Epilepsy Dataset. OpenNeuro ds004100 (2022). Companion: Bernabei JM, et al. Quantitative approaches to guide epilepsy surgery from intracranial EEG. Brain 2023;146(6):2248–2258.

Catalog citation block

Bernabei JM, Li A, Revell AY, Smith RJ, Gunnarsdottir KM, Ong IZ, Davis KA, Sinha N, Sarma S, Litt B. HUP iEEG Epilepsy Dataset. OpenNeuro ds004100 (2022). doi:10.18112/openneuro.ds004100.v1.1.3.

Methods/companion paper: Bernabei JM, Li A, Revell AY, Smith RJ, Gunnarsdottir KM, Ong IZ, Davis KA, Sinha N, Sarma SV, Litt B. Quantitative approaches to guide epilepsy surgery from intracranial EEG. Brain 2023;146(6):2248—2258. doi:10.1093/brain/awad007.

Prior work using subsets of these data: Kini LG, Bernabei JM, et al., Brain 2019; Bernabei JM, Sinha N, et al., Brain 2022.

Download and access

Field	Value
Official source	https://openneuro.org/datasets/ds004100
Access class	open
Size	TODO(confirm)
Format	edf
BIDS	true
Mirrors	NEMAR ds004100

Source & access

• Repository: OpenNeuro — https://openneuro.org/datasets/ds004100 (NEMAR mirror: https://nemar.org/dataexplorer/detail?dataset_id=ds004100).
• Accession / version: ds004100, snapshot v1.1.3 (initial public release 2022-09-21).
• DOI: doi:10.18112/openneuro.ds004100.v1.1.3.
• License: CC0 (public-domain dedication; no attribution legally required, citation requested).

Data structure

• Format: BIDS (BIDS version 1.4.0), converted with MNE-BIDS. Signals are stored as EDF; metadata as TSV (participants.tsv, *_channels.tsv, *_electrodes.tsv) and JSON sidecars.
• Note: OpenNeuro stores the dataset with git-annex/DataLad.

License

• Name: CC0
• Note (verbatim from the directory): public-domain dedication; citation requested
• Snippets on this site: allowed — the site may ship short derived snippets and precomputed products from this dataset, labelled as derivatives (re-referenced, filtered, cropped) with citation, license and DOI.

Acquisition

Field	Value
Device	clinical iEEG — subdural grids/strips plus depths (ECoG) or depth electrodes only (SEEG); patient-specific montages
Device class	intracranial
Channels	TODO(confirm) (patient-specific montages; no fixed channel count (NEMAR’s “59 channels” is a single-subject summary); value from the catalog prose (registry silent))
Sampling rate (Hz)	256 (registry value: raw 256-1024)
Online filters	not documented in the BIDS sidecars
Reference	not documented in the BIDS sidecars
Mains frequency (Hz)	60
Paradigms	clinical-monitoring, ictal, interictal
Participants	58
Population	58 adults 16–59 with drug-resistant focal epilepsy; SEEG 38, ECoG 20; resection 29 / laser ablation 29
Clinical groups	epilepsy
Sessions	1
Durations	clips: task-ictal (epileptologist-marked seizures) and task-interictal (~5-min artifact-minimized clips); mean 5.6 ± 1.5 seizures plus one interictal clip per patient

Acquisition

• Recording type: clinical intracranial EEG (iEEG) — ECoG (subdural grids, strips, depths) or SEEG (depth electrodes only), recorded during inpatient epilepsy monitoring.
• Channels: patient-specific montages; there is no fixed canonical channel set (each subject carries its own contact list with per-channel type, pathology, and coordinate metadata). The NEMAR-displayed “59 channels” is a single-subject/mirror summary, not a dataset-wide constant.
• Electrode localization: contacts mapped to MNI space (README states ICBM152 MNI); per-subject *_electrodes.tsv / *_coordsystem.json sidecars accompany each montage.
• Sampling rate: heterogeneous across subjects (the raw recordings span roughly 256—1024 Hz).
• Power line / reference: USA 60 Hz mains; online reference/filters are not documented in the BIDS sidecars.
• Recording length: organized as discrete clips (commonly ~5-minute interictal segments plus seizure clips) rather than continuous multi-day records; the companion paper reports a mean of 5.6 ± 1.5 seizures plus one interictal clip per patient for ds004100 (Bernabei et al., Brain 2023).

Participants

58 adult epilepsy-surgery patients (confirmed from the released participants.tsv and README; the NEMAR mirror’s “57” reflects a stale index, not the data). Demographics computed directly from participants.tsv:

• Sex: 31 F, 27 M.
• Age: range 16—59 years, mean 34.3 (all participants are adults at implant).
• Implant type: 38 SEEG, 20 ECoG.
• Therapy: 29 resection, 29 ablation (laser).
• Surgical target: temporal (25), mesial temporal lobe / MTL (17), frontal (10), insular (2), medial frontal lobe / MFL (2), parietal (1), frontal-parietal / FP (1).
• Other clinical fields: Engel outcome class, binary outcome (S = seizure-free / F = not free), handedness, lesion status (lesional / non-lesional), and age at seizure onset are provided per subject. There is no healthy-control group — all subjects are clinical (drug-resistant focal epilepsy).

Tasks / conditions

Data are split into two BIDS “task” labels: task-ictal (board-certified-epileptologist-marked seizure events) and task-interictal (artifact-minimized non-seizure clips). There are no behavioral paradigms or stimuli — this is passive clinical monitoring. Sessions are organized as one session with multiple scans/runs per subject.

Used on this site

L0.1, L5.1

Loader

Fetches only the stated subset. Recorded loader: DataLad (OpenNeuro ds004100).

# OpenNeuro serves ds004100 as a DataLad/git-annex repository; fetch one subject only.
datalad install https://github.com/OpenNeuroDatasets/ds004100.git
cd ds004100
datalad get sub-001   # replace with the subject id you need

Caveats worth knowing

• No fixed channel count; sampling rate varies 256–1024 Hz across patients.
• NEMAR’s “57 subjects”/“59 channels” summaries are stale.

Notable caveats

• No fixed channel count. Montages are patient-specific. (NEMAR’s “59 channels” should not be taken as dataset-wide.)
• Subject-ID mismatch warning. The NEMAR mirror flags that participant IDs in participants.tsv may not line up with filename IDs in some tooling (an “RID” vs “HUP/HUD” prefix issue); verify the ID join before batch-processing.
• Coordinate space. The released electrodes are in MNI (ICBM152 per the README).
• No AED/medication detail. Anti-epileptic-drug status is not provided in participants.tsv.

Related datasets

• Open multimodal iEEG-fMRI dataset, naturalistic film

# Loader recorded in the catalog: DataLad (OpenNeuro ds004100)
# TODO(confirm) the exact call and the subset to fetch. Official source: https://openneuro.org/datasets/ds004100