Discover datasets

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

126

Size

3.81 GB

Version

vv1.0.0

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

72

Size

127 GB

Version

vv1.0.0

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

25

Size

21.6 GB

Version

vv1.0.0

Updated

Jul 10, 2026

Participants

31

Channels

62 (10-10)

Citations

8

HED

v8.4.0

Size

7.34 GB

Version

vv1.0.1

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 standardized 10-20 montage, annotated with target and non-target visual stimulus events. This derivative dataset is part of the largest public P300 BCI dataset collection, and is optimized for machine learning applications in BCI research.

Participants

19

Channels

32 (10-10)

HED

v8.4.0

Size

1.46 GB

Version

vv1.0.1

Updated

Jul 10, 2026

A multi-day, high-quality EEG dataset for motor imagery brain-computer interface research comprising 51 healthy subjects performing left and right hand motor imagery tasks across three sessions, with an additional 11 subjects performing 3-class motor imagery (left hand, right hand, foot). The dataset includes 59 EEG channels sampled at 1000 Hz with standardized 10-05 electrode montage, totaling 39,600 trials (51 subjects × 3 sessions × 200 trials for 2-class + 11 subjects × 3 sessions × 300 trials for 3-class) with visual and auditory cues. This resource is designed to support the development and benchmarking of motor imagery BCI algorithms and classifiers.

Participants

51

Channels

59 (10-10)

Citations

29

HED

v8.4.0

Size

63.4 GB

Version

vv1.0.1

Updated

Jul 10, 2026

HEFMI-ICH is a hybrid EEG-fNIRS motor imagery dataset designed for brain-computer interface applications in intracerebral hemorrhage rehabilitation. The dataset comprises 37 participants (17 healthy controls and 20 ICH patients) performing 2-class hand motor imagery tasks (left and right hand grasping) across 1-6 sessions per subject. With 32-channel EEG recordings at 256 Hz and approximately 3,330 trials, this dataset supports the development and evaluation of BCI systems for clinical stroke rehabilitation.

Participants

37

Channels

32 (10-10)

Citations

4

HED

v8.4.0

Size

2.58 GB

Version

vv1.0.1

Updated

Jul 10, 2026

A derivative 4-class code-VEP brain-computer interface dataset processed from the original Zenodo dataset (10.5281/zenodo.8255618), 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 from 12 healthy participants. Burst c-VEP achieved superior accuracy of 95.6% (with 52.8s calibration) and 90.5% (with 17.6s calibration), compared to m-sequence performance of 85.0% and 71.4% respectively. CNN-based decoding with 250ms sliding windows enabled these results. This derivative dataset optimizes stimulus design for reactive BCI applications while maintaining visual comfort, with reduced amplitude (40%) showing minimal accuracy loss (94.2%) while substantially improving user experience.

Participants

12

Citations

20

HED

v8.4.0

Size

148 MB

Version

vv1.0.0

Updated

Jul 10, 2026

This dataset comprises electroencephalographic recordings from 12 healthy participants performing a reactive brain-computer interface task using code-modulated visual evoked potentials (c-VEP). The study introduces an innovative 'Burst c-VEP' paradigm employing short bursts of aperiodic visual flashes at 2-4 Hz, designed to optimize classification performance while reducing calibration requirements and improving visual comfort. Data were collected using a 32-channel BrainProducts LiveAmp system at 500 Hz sampling rate across a factorial design manipulating stimulus pattern (burst vs. m-sequence) and amplitude (100% vs. 40% depth modulation).

Participants

12

Citations

20

HED

v8.4.0

Size

142 MB

Version

vv1.0.0

Updated

Jul 10, 2026

This dataset comprises EEG recordings from 12 healthy participants performing a 4-class code-modulated Visual Evoked Potential (c-VEP) brain-computer interface task. The study compares novel burst c-VEP sequences with traditional m-sequences at two stimulus amplitude depths (100% and 40%) to optimize classification performance while minimizing calibration requirements and enhancing user experience. Data were acquired at 500 Hz using a 32-channel BrainProducts system and processed with convolutional neural network architectures, achieving classification accuracies up to 95.6%.

Participants

12

Citations

20

HED

v8.4.0

Size

148 MB

Version

vv1.0.0

Updated

Jul 10, 2026