Academic-led infrastructure to scale free open neuroimaging data

Brain models need scale. The scans already exist.

Scaling Neuro helps academic labs easily share their already collected fMRI, eventually unlocking the world's largest trainable neuroimaging dataset. Point it at a DICOM folder and it finds the EPI series, removes identifying metadata locally, and syncs the deidentified DICOMs to a shared archive.

Install neuro-sync Request archive access

From the researchers behind MindEye, CortexMAE, and Brainmarks

The scale opportunity

If 32 centers join our effort, we'd share more data than OpenNeuro / HCP / UKBB combined in 1 year.

Aggregating all publicly available fMRI yields ~32K hours of scanning. >1000 hours of fMRI is routinely collected yearly at many universities. Thus, if just 32 universities join Scaling Neuro, we can share over 32K hours of fMRI yearly.

Sources and assumptions

The public baseline is the estimate across freely accessible archives in NeuroAI for AI Safety. The network bar is 32 centers × 1,000 hours/year. The shareable total would be lower after institutional, consent, and privacy review. The dotted curve extends the data-scaling trend reported in CortexMAE; it illustrates the opportunity rather than making a quantitative performance projection.

All public fMRI released ~32,000 hours

Across freely accessible archives like OpenNeuro, UKBB, and HCP

32 centers for one year ~32,000 hours

With enough centers on board, we can share more fMRI than all public archives ever released

How it works

Share approved fMRI from your lab.

01 · Workstation

Find

Walk one folder recursively and identify functional EPI time series from DICOM metadata.

02 · Local privacy

Protect

Remove direct identifiers, dates, paths, free text, and unsafe private metadata locally. Preserve Pixel Data.

03 · Shared archive

Sync

Upload one deidentified DICOM archive per EPI series with resumable transfers and content hashes.

04 · Participating labs

Use

Participating labs get access to all the DICOMs without restrictions.

Functional EPI only. By only uploading raw functional scans that have identifiable metadata removed locally, we avoid all the friction associated with uploading brain scans. No task labels mean your study can't be scooped. No structurals mean no worrying about defacing. No BIDS, so no need to do anything other than point a script to a folder of DICOMs. *Note: Scaling Neuro serves a very different purpose to OpenNeuro; we hope you upload your raw data here and then upload standardized data over there!

Prepare for open sharing

One terminal command No Python or Docker DICOM in, DICOM out

# macOS or Linux
$ curl -fsSL https://scalingneuro.org/install.sh | sh
✓ neuro-sync 0.5.0 downloaded and SHA-256 verified

# One command for each approved scanner export
$ neuro-sync /path/to/dicom-folder
✓ One-time lab registration stays in the terminal
✓ Interrupted work resumes from the same folder command

$ neuro-sync ./new_session_dicoms

Reading DICOM headers          complete
Functional EPI series          4
Other series left local        11
Deidentified DICOM archives    uploaded

Status: complete
Receipt: 4 EPI series safely stored
Report: ~/.local/share/neuro-sync/reports/...

source    original DICOMs stay unchanged on the lab workstation
select    only functional EPI series pass the metadata classifier
identity  direct identifiers, dates, paths, and unsafe text stay local
pixels    scanner-native EPI Pixel Data is preserved
archive   one deterministic dicom.tar.zst per accepted series
transfer  resumable upload with size and SHA-256 verification
compute   none; labs preprocess downloaded DICOMs themselves
hold      non-EPI, unsafe, malformed, and ambiguous series stay local

macOS, Windows, and Linux · functional EPI DICOM sync

Access data

Join the open data effort.

Archive access is available to researchers whose lab agrees to participate in this shared functional MRI effort. Submit the form once to receive a personal access token.

Deidentified functional EPI DICOM archives
Acquisition metadata retained for independent preprocessing
No required cloud compute, conversion, or analysis service

Access granted

Save this token now.

It is shown once. Treat it like a password.

Save it in your shell, then list the archive:

export SCALING_NEURO_ACCESS_TOKEN='PASTE_TOKEN'
curl -H "Authorization: Bearer $SCALING_NEURO_ACCESS_TOKEN" \
  https://scalingneuro.org/v1/archive

Use any listed download_url to download a series:

curl -L -H "Authorization: Bearer $SCALING_NEURO_ACCESS_TOKEN" \
  DOWNLOAD_URL -o series.dicom.tar.zst
shasum -a 256 series.dicom.tar.zst