FOMO45K: Brain MRI Dataset for Large-Scale Self-Supervised Learning with Clinical Data

Dataset paper preprint: A large-scale heterogeneous 3D magnetic resonance brain imaging dataset for self-supervised learning.
https://arxiv.org/pdf/2506.14432.

This repo contains FOMO45K containing ~45K brain MRI scans. This dataset is a subset of FOMO60K that formed the basis for the FOMO25: Foundation Model Challenge for Brain MRI hosted at MICCAI 2025.

See fomo25.github.io and this paper for more information about the challenge.

We also provide FOMO260K, a larger version with over 260K scans without co-registration and skull-stripping.

Description

FOMO45K is a large-scale dataset of brain MRI scans, including both clinical and research-grade scans. The dataset includes a wide range of sequences, including T1, MPRAGE, T2, T2*, FLAIR, SWI, T1c, PD, DWI, ADC, and more.

The dataset consists of

• 9,490 subjects
• 11,967 sessions
• 46,149 scans.

For further details about the dataset, including the preprocessing, please consult the dataset paper preprint.

The FOMO-MRI Dataset Collection

FOMO45K is one of four related datasets in the FOMO-MRI collection. FOMO300K is the full superset; every other dataset in the collection — including this one — is a subset of it. The variants exist to offer different trade-offs between size, access requirements, and preprocessing.

Dataset	Scans	Access	Description
FOMO300K	306,207	Gated (auto-approved)	The full superset across the collection. Gated because some constituent datasets require Data Use Agreements.
FOMO260K	260,927	Open (CC-BY-NC-SA)	The freely accessible subset of FOMO300K. No login or access request required.
FOMO50K	49,193	Gated (auto-approved)	A co-registered, skull-stripped, and/or defaced subset of FOMO300K.
FOMO45K (this dataset)	46,149	Open (CC-BY-NC-SA)	The freely accessible subset of FOMO50K. No login or access request required.

⚠️ Do not combine datasets from this collection. Because each dataset is a subset of FOMO300K, combining them will result in duplicated scans.

Format

All data is provided as NIfTI-files. The dataset is provided as a collection of datasets, each within its own folder PTXYZ_DatasetName (e.g., PT005_IXI). All data has been standardized and preprocessed (including skull stripped, RAS reoriented, co-registered) with the following format:

-- PT005_IXI
   |-- sub_102
       |-- ses_1
           |-- t1.nii.gz
-- PT006_SOOP
   |-- sub_44
       |-- ses_1
           |-- t1.nii.gz

Sessions with multiple scans of the same sequence are named sequence_x.nii.gz. Sessions where the sequence information was not available are named scan_x.nii.gz.

The dataset has been collected from the following public sources: BraTS-GEN, MSD-BrainTumor, IXI, NKI, SOOP, NIMH, DLBS, IDEAS, ARC, MBSR, UCLA, QTAB, AOMIC ID1000.

Metadata Files

The dataset includes the following metadata files, available both in the main folder and in each individual dataset folder:

• participants.tsv: Contains demographic and clinical information including age, gender, handedness, and subject group (e.g., control, specific diagnosis)
• mapping.tsv: Links the files in FOMO45K to the original data source scans
• mri_info.tsv: Includes MRI acquisition information

Additionally, the main folder contains:

• FOMO45K_260K_mapping.tsv: Maps the files that are present in both the 45K and 260K versions of the dataset

Citation

Users must cite the following paper when using the FOMO45K dataset:

@article{Cerri2026large,
  title={A large-scale heterogeneous 3D magnetic resonance brain imaging dataset for self-supervised learning},
  author={Cerri, Stefano and Munk, Asbj{\o}rn and Llambias, Sebastian N{\o}rgaard and Ambsdorf, Jakob and Machnio, Julia and Nersesjan, Vardan and Hedeager Krag, Christian and Liu, Peirong and Rocamora Garc{\'\i}a, Pablo and Mehdipour Ghazi, Mostafa and Boesen, Mikael and Benros, Michael Eriksen and Iglesias, Juan Eugenio and Nielsen, Mads},
  journal={arXiv preprint arXiv:2506.14432},
  year={2026},
  url={https://arxiv.org/abs/2506.14432}
}

In addition, users must comply with all attribution requirements of the constituent datasets included in FOMO45K, as specified in the Usage Notes of the paper.