FOMO300K: 300,000 Heterogeneous 3D Brain MRI Scans

FOMO300K consists of approximately 300,000 3D brain MRI scans compiled by FOMO-MRI for large-scale self-supervised learning. The dataset integrates a heterogeneous mix of clinical and research-grade imaging sequences including T1, T2, FLAIR, and DWI. It was released in 2025 to support the development of medical imaging foundation models.

Use Cases

• Self-supervised pre-training of foundation models using 3D sequences like T1 and FLAIR
• Zero-shot classification of brain images across modalities such as SWI and PD
• Feature extraction for clinical diagnostic support using 3D magnetic resonance data

Strengths

• Scale of 300,000 3D MRI volumes
• Sequence diversity including T1, T2, FLAIR, SWI, and DWI
• Inclusion of both clinical and research-grade scans

Limitations

• Primarily unlabeled data intended for self-supervised learning rather than supervised classification
• High heterogeneity in acquisition parameters across clinical sources may introduce noise

Provenance

Source

FOMO-MRI (Arxiv 2506.14432)

Freshness

Last updated March 2026

Geography

United States