Dataset catalog
🗂️ Public Medical Segmentation Datasets
Official access · dataset splits · labels · reuse terms · papers
🧭 Each card is a starting point for finding official data, understanding what is labeled, and citing the publication.
How to read the split column: counts and label status reflect the source or challenge release named on each page. Original challenge access can differ from later archive releases; unclear details are identified explicitly.
🧠 Brain and neuro-oncology
| Dataset and institutional origin | Task and modality | Official split and annotation status | Publication |
|---|---|---|---|
| BraTS 2015 🏛️ CBICA, University of Pennsylvania · multi-institutional cohort | Glioma subregions · T1, T1ce, T2, FLAIR MRI | Historical release; use the edition-specific source for exact files and split | |
| BraTS 2019 🏛️ CBICA, University of Pennsylvania · multi-institutional cohort | Glioma subregions · T1, T1ce, T2, FLAIR MRI | Training labels; validation and test were challenge evaluation cohorts | |
| BraTS 2021 🏛️ CBICA/UPenn lead; RSNA, ASNR, MICCAI · multi-institutional | Glioma subregions · T1, T1ce, T2, FLAIR MRI | 1,251 labeled training cases; validation/test challenge references hidden | |
| UCSF-BMSR 🏛️ University of California, San Francisco | Brain metastases · multiparametric MRI | 461 exams advertised for download; paper describes a 99-exam challenge test set | |
| BraTS-PEDs 🏛️ CBTN, CONNECT, DIPG Registry · pediatric sites | Pediatric high-grade glioma · multiparametric MRI | 257 labeled train; 91 image-only validation; 109 hidden test | |
| BraTS-MEN 2023 🏛️ 6 centers: Duke, Jefferson, Yale, Missouri, UCSF, Penn | Meningioma subregions · T1, T1ce, T2, FLAIR MRI | 1,000 train; 141 validation; 283 private test; release contents can differ | |
| BraTS-MEN-RT 🏛️ 7 centers: Duke, UCSF, SUNY Upstate, UW, Missouri, KCL, UCSD | Radiotherapy GTV · post-contrast T1 MRI | 500 labeled train; 70 image-only validation; 180 private test | |
| CFB-GBM 🏛️ Centre François Baclesse, France · single-center | Longitudinal glioblastoma · MRI, CT, RT dose/structures | TCIA collection; no official train/validation/test split |
🩻 Abdominal segmentation
| Dataset and institutional origin | Task and modality | Official split and annotation status | Publication |
|---|---|---|---|
| AMOS 2022 🏛️ HKU, CUHK-Shenzhen/SRIBD, Longgang hospitals · multicenter | 15 abdominal organs · CT and MRI | Challenge-time test labels withheld; later Zenodo release describes a labeled archive | |
| BTCV 🏛️ Vanderbilt University Medical Center | 13 abdominal organs · CT | 30 labeled development cases; 20 test cases without public labels |
🧠 Stroke imaging
| Dataset and institutional origin | Task and modality | Official split and annotation status | Publication |
|---|---|---|---|
| AISD 🏛️ Research teams: BUPT, Deepwise, Jinling/Nanjing Univ., Peking Univ., HKU; acquisition site not stated | Acute ischemic infarct · non-contrast CT | 345 train/validation combined; 52 test; test-mask availability not clearly itemized |
🧾 What each dataset page includes
- 🔗 Official dataset, challenge, or institutional access links
- 🗂️ Organizer-defined splits, when documented
- 🏷️ Label definitions and whether validation/test references are public
- ⬇️ Download routes and access requirements
- 🔒 License or reuse terms stated by the source
- 📚 Publication venue, year, volume/pages or article number, DOI, and preprint links when available
Source rule: if the official source does not specify a split, license, or label status, the card says so rather than filling the gap with an unofficial estimate. No images, masks, or paper PDFs are hosted here.
Sources checked: 27 October 2026.