Skip to content
MRI Brain

MGB Clinical Brain MRI Repository

Repository of clinical brain scans with real-world variability

15,926 uncurated, skull-stripped clinical brain MRI scans of 1,384 subjects taken straight from the Massachusetts General Hospital PACS, with mixed sequences, resolutions and orientations plus age and gender. The "MGH dataset" of the SynthSR paper. Access after PI approval and a data use agreement.

Overview

This repository holds routine clinical brain MRI from the picture archiving and communication system (PACS) of Massachusetts General Hospital, part of Mass General Brigham. It is the "MGH dataset" that Iglesias and colleagues used to test SynthSR, a network that turns clinical scans of any contrast, orientation and resolution into synthetic 1 mm isotropic MPRAGE images for 3D morphometry. The data are meant for work on methods that must cope with uncontrolled clinical acquisitions, such as synthesis, super-resolution and segmentation of heterogeneous scans.

Composition

The release contains 15,926 scans from 1,384 subjects as compressed NIfTI files, with a spreadsheet giving each subject's age and gender. The paper describes its subjects as patients with neurology visits and memory complaints at MGH, who are not expected to have large lesions such as tumors or strokes. Each subject has a different number of scans with different sequences. The paper's own analysis used a filtered subset of 9,146 scans after dropping 4D series such as diffusion and scans with an intracranial volume under 1.1 liters.

Acquisition

All scans were acquired during routine clinical care between 2014 and 2026, according to the data use license. Sequences, resolutions, orientations, slice spacing and thickness vary from session to session. The paper shows examples of axial, sagittal and coronal turbo spin echo T1-weighted, T2-weighted and FLAIR scans with 4 to 6 mm slice spacing, and notes that sessions also include angiography and diffusion. Scanner vendors and field strengths are not reported.

Annotations

There are no manual labels. Every scan was skull stripped with SynthSeg+, and the data include all scans that SynthSeg could process.

Known limitations

The collection is uncurated on purpose: many scans are unusable because of a limited field of view, heavy noise or a non-structural sequence. Dates and DICOM headers were removed and ages of 90 and over are aggregated. Scan-level metadata such as sequence type is not described. Access needs approval by the principal investigator, a data use agreement with Mass General Brigham, and is limited to countries allowed under the U.S. Data Security Program.

License and access

Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer

Access
Signed agreement

Sign a data use agreement, often reviewed by the provider

Access page

Data Use License for the MGB Clinical Brain MRI Repository

Access only after the principal investigator approves a request on Zenodo and a data use agreement with Mass General Brigham is executed. Use is limited to non-commercial research such as method development and model training; other uses need written approval. No re-identification, no redistribution, minimum security controls, and access only from countries approved under the U.S. Data Security Program.

Original license text Version read: Version 1.0, March 2026, as shown on the Zenodo record Checked 2026-10-09

What you can do

  • No
  • Not stated
  • Conditional
  • Yes
  • Yes

What you can share

  • No
  • Not stated
  • Not stated

What you must do

  • Yes
  • Share alike No
  • Yes
  • Conditional
  • Manuscript review No
  • Release code No
  • Return results No
  • Delete after use No

Limits

  • Yes
  • Yes

Commercial license: Conditional Uses outside non-commercial research need explicit written approval from the MGB Contracting Office.

Citation

Iglesias JE, et al. Repository of Clinical Brain MRI Scans with Real World Variability, Mass General Brigham (2026). Zenodo. doi:10.5281/zenodo.18701184. And: Iglesias JE, Billot B, Balbastre Y, Magdamo C, Arnold SE, Das S, Edlow BL, Alexander DC, Golland P, Fischl B. SynthSR: A public AI tool to turn heterogeneous clinical brain scans into high-resolution T1-weighted images for 3D morphometry. Science Advances 9(5):eadd3607 (2023). doi:10.1126/sciadv.add3607

Sources

Every number on this page comes from one of these documents. Each chart names the table or page it is taken from. The raw numbers are in stats.csv.