FOMO260K: Large-Scale Brain MRI Dataset for Self-Supervised Learning

FOMO260K is a large-scale dataset of over 260,000 brain MRI scans for self-supervised learning, as indicated by its name. It includes both clinical and research-grade scans across a wide range of sequences such as T1, MPRAGE, T2, and FLAIR. The dataset was created by FOMO-MRI and a related preprint paper was published in 2025.

Use Cases

• Pre-training self-supervised vision models based on the large volume of unlabeled 3D MRI scans.
• Developing domain-specific foundation models for neuroimaging using the heterogeneous mix of clinical and research-grade data.
• Benchmarking medical imaging algorithms on a large-scale dataset with diverse MRI sequences like T1, T2, and FLAIR.
• Studying the transfer of models trained on research-grade scans to clinical data, based on the dataset's inclusion of both types.

Strengths

• Large scale, with over 260,000 scans implied by the dataset name.
• Includes a wide range of MRI sequences such as T1, MPRAGE, T2, T2*, FLAIR, SWI, T1c, PD, and DWI.
• Contains both clinical and research-grade scans, providing heterogeneity for model training.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

FOMO-MRI

Collection Method

Collection of brain MRI scans from clinical and research sources, as described.

Freshness

Last updated 2026-05-06 12:00:27; freshness should be verified.