Discover datasets

A polysomnographic dataset comprising 1,983 overnight sleep recordings from Massachusetts General Hospital, including 994 training subjects with expert annotations and 989 test subjects. The dataset was created for the PhysioNet/Computing in Cardiology Challenge 2018 and contains 13 channels of physiological signals (EEG, EOG, EMG, respiratory, SpO2, ECG) sampled at 200 Hz, with annotations for sleep stages, respiratory events, and arousal types in the training set.

Participants

1983

Channels

6 (other)

Citations

91

Size

401 GB

Version

v1.1.0

Updated

Jul 10, 2026

This dataset comprises scalp EEG recordings from 45 participants during passive listening to a naturalistic 12.4-minute audiobook excerpt (first chapter of Alice's Adventures in Wonderland). The study investigates how the brain constructs hierarchical syntactic structure and generates rapid linguistic predictions during continuous speech comprehension. Each word event is annotated with linguistic predictors including lexical frequency, acoustic properties, and surprisal estimates from n-gram, recurrent neural network, and context-free grammar models.

Participants

45

Channels

60 (other)

Citations

127

Size

3.81 GB

Version

v1.0.0

Updated

Jul 10, 2026

Resting-state 62-ch EEG, 215 subjects, eyes-open/closed blocks. BrainVision actiCHamp 2500Hz. DOI:10.1038/sdata.2018.308

Participants

215

Channels

61 (10-10)

Citations

74

Size

127 GB

Version

v1.0.0

Updated

Jul 10, 2026

64-ch EEG, 95 subjects, 2 sessions, 6 paradigms (13 tasks). BrainAmp 250Hz Easycap 64-ch. DOI:10.1016/j.neuroimage.2022.119666

Participants

95

Channels

64 (10-10)

Citations

28

Size

21.6 GB

Version

v1.0.0

Updated

Jul 10, 2026

BigP3BCI Study P is a P300-based brain-computer interface dataset comprising EEG recordings from 19 healthy subjects across 2 sessions each, using a 9x8 character grid spelling paradigm. The dataset contains 32-channel EEG data sampled at 256 Hz with target and non-target visual stimuli annotated using HED 8.4.0 schema. This derivative dataset is part of the larger BigP3BCI collection, the largest public P300 BCI dataset, and is optimized for machine learning applications in BCI research.

Participants

19

Channels

32 (10-10)

HED

v8.4.0

Size

4.96 GB

Version

v1.0.3

Updated

Aug 18, 2026

This dataset (WBCIC-SHU) contains multi-day EEG recordings from 51 healthy, right-handed, BCI-naive subjects performing a motor imagery brain-computer interface paradigm across three sessions per subject. Participants imagined left-hand, right-hand, or (for a subset of 11 subjects) foot movements in response to visual and auditory cues, yielding 39,600 trials in total. The dataset is intended for benchmarking motor imagery classification algorithms such as CSP, FBCSP, EEGNet, deepConvNet, and FBCNet, and was converted to BIDS format using MOABB.

Participants

51

Channels

59 (10-10)

Citations

34

HED

v8.4.0

Size

154 GB

Version

v1.0.5

Updated

Aug 18, 2026

A 4-target, 2-class code-VEP brain-computer interface dataset from 12 healthy participants comparing burst c-VEP and m-sequence stimulation paradigms at two amplitude depths (100% and 40%). The study evaluates classification performance and user experience using 32-channel EEG recorded at 500 Hz. This derivative dataset is derived from the original Castillos et al. (2023) study and optimizes stimulus design for reactive BCI applications while maintaining visual comfort. CNN-based decoding achieved up to 95.6% accuracy (best-case scenario with 52.8 s of calibration data).

Participants

12

Channels

32 (10-10)

Citations

21

HED

v8.4.0

Size

352 MB

Version

v1.0.2

Updated

Aug 18, 2026

This dataset comprises electroencephalographic recordings from 12 healthy participants performing a reactive brain-computer interface task based on code-modulated visual evoked potentials (c-VEP). The study introduces an innovative 'Burst c-VEP' paradigm using short bursts of aperiodic visual flashes at 2-4 Hz, systematically varying stimulus amplitude (100% vs. 40%) to optimize classification performance while improving visual comfort. Data were collected using a 32-channel BrainProducts LiveAmp system at 500 Hz sampling rate and analyzed with convolutional neural networks for bitwise decoding.

Participants

12

Channels

32 (10-10)

Citations

21

HED

v8.4.0

Size

337 MB

Version

v1.0.2

Updated

Aug 18, 2026

This dataset comprises EEG recordings from 12 healthy participants performing a code-modulated visual evoked potential (c-VEP) brain-computer interface task. The study introduces Burst c-VEP, a novel variant using short bursts of aperiodic visual flashes, and compares its performance to traditional m-sequence stimuli across two amplitude depths (100% and 40%). Data were acquired at 500 Hz using a 32-channel BrainProducts system and include 60 trials per subject with convolutional neural network-based classification achieving 95.6% accuracy.

Participants

12

Channels

32 (10-10)

Citations

21

HED

v8.4.0

Size

352 MB

Version

v1.0.2

Updated

Aug 18, 2026

A code-modulated Visual Evoked Potential (c-VEP) electroencephalography dataset from 12 healthy participants investigating burst c-VEP as a reactive brain-computer interface paradigm. This dataset contains m-sequence c-VEP data at 40% amplitude modulation depth. The original study employed a factorial design manipulating stimulus pattern (burst vs. m-sequence) and amplitude depth (100% vs. 40% modulation) to optimize classification accuracy while improving user visual comfort. Data were acquired at 500 Hz using 32-channel EEG with convolutional neural network classification achieving up to 95.6% accuracy on the full amplitude burst condition.

Participants

12

Channels

32 (10-10)

Citations

22

HED

v8.4.0

Size

340 MB

Version

v1.0.2

Updated

Aug 18, 2026