---

Last updated: 2026-09-11

A maintained, machine-readable catalog of medical AI datasets and benchmarks for training and evaluation. Datasets describe released data; benchmarks separately describe evaluation protocols, metrics, and their constituent datasets.

Resources are organized by the data they release:

• Text Only: clinical and biomedical language resources.
• Image Only: medical images with labels, bounding boxes, masks, or other visual annotations.
• Text-Image Pairs: reports, captions, question-answer pairs, grounded text, and other image-language supervision.

Each dataset is a version-controlled record in datasets/records/, not just a link. Records capture clinical domains and modalities, release year and scale, tasks and capabilities, annotation provenance and expert review, spatial grounding, quality signals, and access or license restrictions. included, candidate, and excluded statuses make the catalog's publication decisions auditable.

Table guide: Data profile is ordered as modality · anatomy · image structure. Annotations describes the released annotation artifact linked to each image or study. License/access cells also state commercial-use status and whether a data-use agreement is required.

Quick Start

Generated from the records in datasets/ and benchmarks/. Do not edit generated content by hand; run python3 scripts/generate_tables.py.

python3 scripts/validate_records.py
python3 scripts/format_records.py --check
python3 scripts/generate_tables.py
python3 scripts/check_links.py --timeout 15

Catalog Status

55 included datasets: 3 text-only, 34 image-only, and 18 text-image pairs. 48 candidate and 1 excluded dataset records are retained for auditability. 5 included benchmarks; 0 candidate and 0 excluded benchmark records are omitted from public tables.

Catalog Navigation

Text Only | Image Only | Text-Image Pairs | Longitudinal Resources | Benchmarks

Datasets By Resource Type

Text Only

Radiology

General Biomedical

Image Only

Radiology

Pathology

Ophthalmology

Ultrasound

Surgery

Text-Image Pairs

Radiology

Pathology

Ophthalmology

Endoscopy

Dermatology

Ultrasound

General Biomedical

Longitudinal Resources

This cross-cutting view includes datasets with an explicit longitudinal sequence or longitudinal-comparison support. Resources remain listed under their primary type above.

7 included datasets with explicit longitudinal sequences or longitudinal-comparison support. These records also remain listed under their primary resource type.

Radiology

Ophthalmology

General Biomedical

Benchmarks

Benchmarks are rendered separately because one benchmark may combine multiple public datasets. A benchmark record links to its companion dataset payload and constituent dataset records without duplicating their provenance or license metadata.

5 included benchmarks, maintained separately from their companion and source datasets.

Capability Coverage

Caveats

Dataset metadata records describe sources, not legal advice. unknown and unclear mean the contributor could not establish an answer from an official source. Credentialed resources may require a data-use agreement or training before download. See DATASET_SCHEMA.md for field definitions and reports/gaps.md for coverage gaps.

Contributing

Please read CONTRIBUTING.md and follow the record workflow. Add one JSON-compatible YAML record per dataset or benchmark, run validation, and commit the regenerated final report.