Skip to content
Open Imaging Index
MRI Breast

Duke-Breast-Cancer-MRI

Dynamic contrast-enhanced magnetic resonance images of breast cancer patients with tumor locations

Pre-operative dynamic contrast-enhanced breast MRI of 922 women with biopsy-confirmed invasive breast cancer from Duke Hospital, with radiologist tumor bounding boxes, 529 radiomic features, clinical, pathology, treatment and outcome data, and breast and fibroglandular tissue masks for a subset.

Go to dataset Edit Report

Overview

Duke-Breast-Cancer-MRI is a retrospective, single-center collection of pre-operative breast MRI from 922 women with biopsy-confirmed invasive breast cancer, seen at Duke Hospital between January 2000 and March 2014. The Mazurowski lab at Duke University School of Medicine assembled it for radiogenomics and outcome prediction work, and The Cancer Imaging Archive hosts it. The images come with a large table of clinical, pathology, treatment and follow-up variables, so the set is used for tumor detection, molecular subtype and receptor status prediction, response to neoadjuvant therapy and recurrence modelling.

Composition

Each patient has one MRI study with four to six MR series: a T1-weighted series without fat suppression, a fat-suppressed gradient echo T1-weighted pre-contrast series and, in most cases, three or four post-contrast phases. Patients were aged 21 to 89 at diagnosis, with a median of 52. The clinical table covers demographics, ER, PR and HER2 status, surrogate molecular subtype, TNM stage, grade and histology, MRI, mammography and ultrasound report findings, surgery, radiation, chemotherapy, endocrine and anti-HER2 therapy, pathologic response, recurrence and follow-up times. A second table holds 529 computer-extracted features of the tumor and fibroglandular tissue.

Acquisition

Axial scans were acquired in the prone position on GE and Siemens scanners at 1.5 T or 3 T. Contrast agents were gadopentetate dimeglumine for most patients, gadobenate dimeglumine for about 29% and gadobutrol for two, with the agent unrecorded for the rest. A table lists the scanner model and acquisition parameters per patient.

Annotations

Eight fellowship-trained breast radiologists drew a 3D bounding box around the biopsied tumor for every patient, working on the pre-contrast, first post-contrast and subtraction images. The 271 earlier cases followed a slightly different procedure from the remaining 651. Later versions added manual breast and fibroglandular tissue masks: DICOM SEG masks for 127 patients, and NRRD breast, fibroglandular tissue and vessel masks for 100 patients that breast radiologists reviewed.

Known limitations

  • Single institution, so external validity is limited.
  • Tumor annotations are bounding boxes only; voxel-level tumor masks are not included.
  • Acquisition parameters vary widely across scanners and years.
  • The Frame of Reference UID was replaced during de-identification and may not be reliable for aligning series.

Cohort

Aggregate numbers from the sources below. Bars are relative to the 922 subjects.

Sex

  • Female 922 100%

Age

mean 52.9± 11.2 · range 21.75 to 89.49
0
100
200
290
20-2930-3940-4950-5960-6970-7980-89

Modality

subjects

  • MRI 922 100%

Condition

subjects, values can overlap

  • Breast cancer 922 100%

Anatomy

subjects

  • Breast 922 100%

Scanner vendor

subjects

  • GE HealthCare 628 68%
  • Siemens Healthineers 294 32%

Field strength

subjects

  • 1.5 T 468 51%
  • 3 T 453 49%

Country

subjects

  • United States 922 100%

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

DICOM images and DICOM SEG masks download with the TCIA Data Retriever. Clinical, annotation box, radiomic feature and file path tables are direct XLSX or CSV downloads; the NRRD breast and fibroglandular tissue masks are Aspera downloads.

Access page

Creative Commons Attribution-NonCommercial 4.0 International

Use, share and adapt the data with credit, but only for non-commercial purposes.

Every item on the TCIA collection page is listed under CC BY-NC 4.0. TCIA also asks users to follow its Data Usage Policy and to cite the dataset DOI.

Original license text Version read: 4.0 Checked 2026-10-07

What you can do

  • No
  • Conditional
  • Yes
  • Yes

What you can share

  • Conditional
  • Conditional
  • Conditional

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

Saha A, Harowicz MR, Grimm LJ, Weng J, Cain EH, Kim CE, Ghate SV, Walsh R, Mazurowski MA (2021). Dynamic contrast-enhanced magnetic resonance images of breast cancer patients with tumor locations [Data set]. The Cancer Imaging Archive. https://doi.org/10.7937/TCIA.e3sv-re93. Publication: Saha A, et al. A machine learning approach to radiogenomics of breast cancer: a study of 922 subjects and 529 DCE-MRI features. British Journal of Cancer 119(4), 508-516 (2018). https://doi.org/10.1038/s41416-018-0185-8

All numbers

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

MeasureBreakdownValueSource
Subjectstotal 922saha2018
Table 2
Subjectscondition=breast_cancer
invasive breast cancer
922saha2018
Materials and methods, Patient population
Subjectsmodality=MR 922tcia-duke-breast-cancer-mri
Data Access (Version 3)
Subjectssex=female 922saha2018
Materials and methods, Patient population
Subjectsanatomy=breast 922tcia-duke-breast-cancer-mri
Data Access (Version 3)
Subjectscountry=US
Duke Hospital
922tcia-duke-breast-cancer-mri
Detailed Description, Population
Studiestotal 922tcia-duke-breast-cancer-mri
Data Access (Version 3)
Scanstotal
DICOM series including 127 SEG series
5,161tcia-duke-breast-cancer-mri
Data Access (Version 3)
Scansmodality=MR 5,034tcia-duke-breast-cancer-mri-digest
Modality
Imagestotal
DICOM files including 127 SEG objects
773,253tcia-duke-breast-cancer-mri
Data Access (Version 3)
Median agetotal 52.3saha2018
Table 2
Minimum agetotal 21.8saha2018
Table 2
Maximum agetotal 89.5saha2018
Table 2
Mean agetotal
age at diagnosis
52.9tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Age SDtotal
age at diagnosis
11.2tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=20-29
age at diagnosis
17tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=30-39
age at diagnosis
90tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=40-49
age at diagnosis
275tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=50-59
age at diagnosis
290tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=60-69
age at diagnosis
184tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=70-79
age at diagnosis
60tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsage=80-89
age at diagnosis
6tcia-duke-breast-cancer-mri-clinical
Date of Birth (Days)
Subjectsvendor=ge 628tcia-duke-breast-cancer-mri-clinical
Manufacturer
Subjectsvendor=siemens 294tcia-duke-breast-cancer-mri-clinical
Manufacturer
Subjectsfield_strength=1.5 468tcia-duke-breast-cancer-mri-clinical
Field Strength (Tesla)
Subjectsfield_strength=3
one more subject reports 2.8936 T
453tcia-duke-breast-cancer-mri-clinical
Field Strength (Tesla)
Scansvendor=ge
MR series
3,485tcia-duke-breast-cancer-mri-digest
Modality, Manufacturer
Scansvendor=siemens
MR series
1,549tcia-duke-breast-cancer-mri-digest
Modality, Manufacturer

Sources

The keys used in the table above.