Breast imaging datasets
This page lists the breast imaging datasets that we keep locally. Each entry gives the source, the size of our copy, and the labels that it has. The counts are for our copy, which can be a subset of the published dataset. Each dataset has its own page with an example breast and its labels.
| Dataset | Images | Source | Labels |
|---|---|---|---|
| EMBED (subset) | 27,438 | Emory University, USA | Density, BI-RADS, pathology, lesion boxes |
| VinDr-Mammo | 20,000 | Two hospitals, Vietnam | Density, BI-RADS, finding boxes |
| M-SYNTH | 14,760 | FDA VICTRE simulation | Density, lesion truth and position |
| CMMD | 5,202 | China | Benign or malignant, molecular subtype |
| TOMPEI-CMMD | 1,385 annotations | Annotations for CMMD | Lesion outlines |
| CBIS-DDSM (mass) | 1,592 | Scanned film, USA | BI-RADS, pathology, mass masks |
| CDD-CESM | 1,004 | Egypt | BI-RADS, lesion outlines, reports |
| DMID | 510 | India | Lesion masks, reports |
| INbreast | 412 | Porto, Portugal | Density, BI-RADS, lesion outlines |
EMBED, VinDr-Mammo and INbreast do not allow their images to be redistributed, so their pages show a schematic of the label types instead of a real image. The figures are made by scripts/make_dataset_figures.py.