Skip to content
Open Imaging Index
Mammography Breast

CBIS-DDSM

Curated Breast Imaging Subset of the Digital Database for Screening Mammography

Curated, DICOM-converted subset of the 1990s DDSM scanned-film mammograms: 1,566 women, 3,103 CC and MLO mammograms with mass and calcification ROIs, BI-RADS descriptors, pathology labels and a fixed train/test split.

Go to dataset Edit Report

Overview

CBIS-DDSM is a cleaned-up and standardised selection from the Digital Database for Screening Mammography (DDSM), a collection of scanned film mammograms assembled in the 1990s. A Stanford group released it through The Cancer Imaging Archive in 2016-2017 so that mammography CAD methods could be compared on one public benchmark. It is used for mass and calcification classification, lesion detection and mass segmentation.

Composition

The collection covers 1,566 participants. The data descriptor reports 891 mass cases and 753 calcification cases; the public description tables list 3,568 abnormalities on 3,103 full mammograms in craniocaudal (CC) and mediolateral oblique (MLO) views. Every abnormality carries a pathology label (malignant, benign, or benign without callback), a BI-RADS assessment, a subtlety score, breast density and shape, margin, type or distribution descriptors. About a fifth of the cases form a test set, split separately for masses and calcifications and stratified by BI-RADS category. Besides the full images, the download contains ROI masks and square crops around each abnormality. In the DICOM headers every breast view has its own patient id, so TCIA's counters show 6,671 "patients".

Acquisition

DDSM mammograms came from Massachusetts General Hospital, Wake Forest University School of Medicine, Sacred Heart Hospital and Washington University in St Louis, digitised on several film scanners. The curators decompressed the obsolete lossless JPEG files, mapped pixel values to standardised optical density and stored the result as 16-bit DICOM.

Annotations

A trained mammographer reviewed questionable mass cases and removed 254 images in which the mass was not clearly visible. Mass outlines were refined with a level-set segmentation started from the original DDSM contours; calcification ROIs keep the original, coarse outlines.

Known limitations

  • Scanned film from decades ago; no digital mammography and no normal cases.
  • Calcification ROIs are imprecise.
  • Age and other demographics are not in the description tables, although DDSM recorded patient age.
  • Train and test are split per abnormality, so a few participants appear in both.

Cohort

Aggregate numbers from the sources below. Bars are relative to the 1,566 subjects.

Condition

subjects, values can overlap

  • Breast cancer 752 48%

Anatomy

subjects

  • Breast 1,566 100%

Split

images

  • Training 2,458 79%
  • Test 645 21%

View

subjects

  • Mediolateral oblique 1,473 94%
  • Craniocaudal 1,312 84%

License and access

Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer

Access
Open download

Download without an account

Access page

Creative Commons Attribution 3.0 Unported

Older version of CC BY. Use, share and adapt the data for any purpose, including commercial use, as long as you credit the creators.

TCIA lists the images and the four description CSV files under CC BY 3.0. Users must also agree to the TCIA Data Usage Policy, which says users must "Not use the requested datasets, either alone or in concert with any other information, to identify or contact individual participants" and must cite the dataset DOI (https://www.cancerimagingarchive.net/data-usage-policies-and-restrictions/).

Original license text Version read: 3.0 Checked 2026-10-07

What you can do

  • Yes
  • Yes
  • Yes
  • Yes

What you can share

  • Yes
  • Yes
  • Yes

What you must do

  • Yes
  • Share alike No
  • No
  • No
  • Manuscript review No
  • Release code No
  • Return results No
  • Delete after use No

Limits

  • No
  • Location limits No

Citation

Sawyer-Lee R, Gimenez F, Hoogi A, Rubin D. Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM) [Data set]. The Cancer Imaging Archive (2016). https://doi.org/10.7937/K9/TCIA.2016.7O02S9CY. Lee RS, et al. A curated mammography data set for use in computer-aided detection and diagnosis research. Sci Data 4, 170177 (2017).

All numbers

Every number on this page, as stored in stats.csv, with its source.

MeasureBreakdownValueSource
Subjectstotal
Actual participants; DICOM metadata shows 6671 patient ids because each id encodes breast and view
1,566tcia-cbis-ddsm
Subjectsanatomy=breast 1,566tcia-cbis-ddsm
Imagestotal
Full-mammogram DICOM series referenced in the four CSVs; 71 breast views appear twice, once as mass and once as calcification case
3,103cbis-ddsm-case-csvs
Imagessplit=train 2,458cbis-ddsm-case-csvs
Imagessplit=test
The split is per abnormality; 31 participants have images in both splits
645cbis-ddsm-case-csvs
Imagesview=CC 1,460cbis-ddsm-case-csvs
Imagesview=MLO 1,643cbis-ddsm-case-csvs
Imagesview=CC;split=train 1,160cbis-ddsm-case-csvs
Imagesview=CC;split=test 300cbis-ddsm-case-csvs
Imagesview=MLO;split=train 1,298cbis-ddsm-case-csvs
Imagesview=MLO;split=test 345cbis-ddsm-case-csvs
Subjectsview=CC
Participants with at least one image in this view
1,312cbis-ddsm-case-csvs
Subjectsview=MLO 1,473cbis-ddsm-case-csvs
Imagescondition=breast_cancer
Images with at least one abnormality of pathology MALIGNANT
1,375cbis-ddsm-case-csvs
Subjectscondition=breast_cancer
Participants with at least one abnormality of pathology MALIGNANT
752cbis-ddsm-case-csvs
Imagescondition=breast_cancer;view=CC 649cbis-ddsm-case-csvs
Imagescondition=breast_cancer;view=MLO 726cbis-ddsm-case-csvs

Sources

The keys used in the table above.