EMBED
The Emory Breast Imaging Dataset contains real screening and diagnostic mammograms from Emory University, USA.
Labels
| Label | Level | Notes |
|---|---|---|
| Breast density | Exam | tissueden, 1-4 (BI-RADS A-D) |
| BI-RADS assessment | Finding | asses |
| Pathology | Finding | Linked biopsy result and severity (path_severity), when there is one |
| Lesion box | Image | ROI_coords on 2D images, drawn by a radiologist |
| Demographics | Patient | Age, race, ethnicity |
| Acquisition | Image | Device, view, screening or diagnostic exam |
Our copy
We have a subset of the open release, because the full release is 2.4 TiB. The subset has two parts:
- Screening exams with the 4 standard views, from one Hologic device model, with one exam for each patient. Each BI-RADS density has up to 1,500 exams.
- All 2D images that have a radiologist lesion box.
27,438 images in all, stored as lossless DICOM.
Source
- Site: AWS Open Data Registry, EMBED_Open_Data on GitHub
- Paper: Jeong et al. 2023, Radiology: Artificial Intelligence. doi:10.1148/ryai.220047
- License: custom data use agreement. Each user must register. Do not share the data, the download link, or images from it.