Overview
The Healthy Brain Network EEG Datasets (HBN-EEG) is a curated collection of high-resolution EEG data from over 3,000 participants aged 5-21 years, contributed by the Child Mind Institute Healthy Brain Network (HBN) project.
Currently, HBN-EEG includes 11 dataset releases in the Brain Imaging Data Structure (BIDS) format, containing EEG and behavioral data from different subjects performing the same tasks. Further dataset releases will be added as the study progresses.
These datasets also contain rich task-event annotations using Hierarchical Event Descriptors (HED) , making them ideal for large-scale analyses and machine-learning inference related to child and adolescent mental health, including advanced meta and mega analyses. Future enhancements will include more data layers such as eye-tracking and personalized EEG features.
Data Description
Contents
- EEG Data: EEG recordings from a set of participants performing multiple tasks (some passive, some task-based with behavioral input).
- Behavioral Responses: Reaction times and accuracy are integrated with EEG events for streamlined analysis in
_events.tsv
files. - Participant Information: Includes demographic information (age, sex, handedness) and psuchopathology dimensions derived from Child Behavior Checklist questionnaires.
- Hierarchical Event Descriptors (HED): Task events are annotated using HED, allowing for flexible and detailed event analysis across different tasks and conditions.
- Dataset Quality Checks: Each dataset release includes quality control checks to ensure data integrity, including checks for data length, sampling frequency, and event marker availability. All inconsistencies are flagged to maintain data quality.
Tasks
The HBN-EEG dataset includes EEG recordings from participants performing six distinct tasks, categorized into passive and active tasks based on user interaction requirements.
Passive Tasks
Resting State: Participants rested with heads on a chin rest, following voice instructions to alternate between eyes-closed and eyes-open periods. During eyes-open periods, they fixated on a central cross displayed on the screen. This baseline condition allows researchers to measure spontaneous brain activity without external task demands.
Surround Suppression: Participants viewed four flashing peripheral disks presented against contrasting backgrounds across two ~3.6-minute runs. This visual paradigm tests how the brain processes competing visual information when foreground and background elements have different contrast properties. Stimulus parameters were synchronized with behavioral recordings to ensure precise event timing.
Movie Watching: Participants watched four themed short movies (‘Despicable Me’, ‘Diary of a Wimpy Kid’, ‘Fun with Fractals’, ‘The Present’) while EEG was recorded. This naturalistic viewing condition captures brain responses to complex, dynamic visual and auditory content, providing insights into neural processing during realistic media consumption.
Active Tasks
Contrast Change Detection: Participants monitored two co-centric flickering grated disks (one left-leaning, one right-leaning) and identified which disk showed increased contrast when a change occurred. They received immediate feedback (smiley or sad face) based on response accuracy. This task measures visual attention, decision-making, and contrast sensitivity.
Sequence Learning: Participants observed and memorized sequences of flashed circles positioned around an invisible circle’s periphery (10 circles for participants ≥8 years, 7 for younger children among 8 or 6 possible targets). After viewing each sequence five times, participants reproduced it using a computer mouse, testing visuospatial memory and motor learning.
Symbol Search: Participants performed a computerized version of the WISC-IV neuropsychological test, searching for target symbols within rows of five search symbols. They worked through 15-row sequences, advancing to new sets after completing all rows. This task assesses visual processing speed, attention to detail, and symbol recognition abilities.
Hierarchical Event Descriptors (HED)
Each dataset release uses Hierarchical Event Descriptors (HED) for comprehensive cross-study annotation. HED’s hierarchical structure enables researchers to analyze data at multiple levels of abstraction, from specific individual events to broad categorical groupings. This powerful annotation system allows researchers to analyze common features of agent actions like button presses collectively, regardless of which hand performed the action, facilitating lateralization studies while maintaining meaningful action-level analysis. Similarly, visual presentations with contrast changes can be grouped together whether they originate from movie stimuli, SSVEP tasks, or contrast detection paradigms, enabling cross-task visual processing research.
The hierarchical organization supports both highly specific task-focused analyses and broad meta-analyses across different experimental paradigms. This structure is particularly valuable for machine learning approaches that can leverage common event features across diverse tasks, enabling researchers to identify shared neural mechanisms underlying different cognitive processes while maintaining the flexibility to drill down into task-specific details when needed.
Psychopathology Dimensions
The datasets include psychometric bi-factors (p-factor, internalizing, externalizing, attention) derived from Child Behavior Checklist questionnaires. These bifactor model-derived dimensions provide privacy-preserving, stable measures of mental health traits that enhance statistical power while protecting participant confidentiality through anonymized factor scores rather than raw responses.
Platform Updates & Future Features
- Personalized Electrode Locations: Future dataset releases will include more precise electrode location files tailored to each participant’s head anatomy.
- Eye-Tracking Data: Upcoming additions will include synchronized eye-tracking data to provide a better understanding of attention and visual behaviors during EEG tasks.
- Lead Field Matrix: Plans include releasing customized lead-field matrices for source imaging analyses.
Number of subjects across releases
Number of subjects with available EEG data runs across HBN-EEG Releases. “Available” flags are determined based on data length, sampling frequency, and availability of event markers (see the HBN-EEG preprint for more details).
Dataset Releases
Below are links to download specific dataset releases. Each release contains EEG data for multiple tasks. The data are structured uniformly to support comparative analyses.
Release 1 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R1
- Total subjects: 136
Release 2 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R2
- Total subjects: 152
Release 3 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R3
- Total subjects: 183
Release 4 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R4
- Total subjects: 324
Release 5 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R5
- Total subjects: 330
Release 6 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R6
- Total subjects: 134
Release 7 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R7
- Total subjects: 381
Release 8 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R8
- Total subjects: 257
Release 9 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R9
- Total subjects: 295
Release 10 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R10
- Total subjects: 295
Release 11 | Download from NEMAR.org
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_R11
- Total subjects: 295
Not for Commercial Use Release | Download from NEMAR.org (NOT YET AVAILABLE, USE S3)
- S3 URI:
s3://fcp-indi/data/Projects/HBN/BIDS_EEG/cmi_bids_NC
- Total subjects: 458
References
If you use this dataset platform, please cite the following publications:
- HBN-EEG data paper: Shirazi et al., bioRxiv 2024 , DOI: 10.1101/2024.10.03.615261.
- HBN project paper: Alexander et al., Sci Data 2017 , DOI: 10.1038/sdata.2017.181.
- Detailed EEG protocols: Langer et al., Sci Data 2017 , DOI: 10.1038/sdata.2017.40.
Acknowledgments
We express our gratitude to the participants and their families for supporting the Healthy Brain Network project. Special thanks to the Child Mind Institute teams for their leadership and support during the data collection process.
HBN-EEG is also made possible by the efforts and computational resources provided by the NeuroScience Gateway team. Funding for preparing the BIDS datasets was partially provided by NIH/NIMH (R01MH125934).
© 2024 The Authors