fidopt Interactive view
Body MR, many structures, segmentation

Duke breast MRI, breast tissues

The fat-saturated T1-weighted MR before contrast of a woman with invasive breast cancer at Duke, with the breasts' fatty tissue, fibroglandular tissue and vessels drawn by hand (Duke-Breast-Cancer-MRI's 3D segmentations, Lew et al. 2024).

10 test casesbreastimage MRresearch only
Look at the images
Read before comparing (8)
  • Pool. Duke-Breast-Cancer-MRI holds the pre-operative DCE MR of 922 patients with invasive breast cancer (Duke Hospital, 2000 to 2014); Lew et al. drew the three tissues on the pre-contrast fat-saturated series of 100 drawn at random, 70 on GE and 30 on Siemens scanners, 40 at 1.5 T and 60 at 3 T.
  • Truth. About 8 hours of drawing a case; every case modified or approved by a breast fellowship-trained radiologist. No agreement between readers is published. Graded expert. The breast's posterior edge is a cut along the chest wall, straight across on some cases (Breast_MRI_112).
  • Labels. The deposit keeps the breast in one file and the fibroglandular tissue and vessels in another; prepare.py writes one map, the breast outside the two being breast_fatty_tissue. A method that outlines the whole breast loses where the fibroglandular tissue and vessels lie. A few voxels of fibroglandular tissue or vessels lie outside the breast mask (none to 1,422 on the test MRs) and stay labelled. Layered files that let the two overlap hold no overlap on the cases built.
  • Cases. On the test MRs the fatty tissue holds 1.4 to 5.5 L, the fibroglandular tissue 100 to 290 mL and the vessels 3 to 26 mL. The study's network scored a Dice of 0.92 for the breast, 0.86 for the fibroglandular tissue and 0.65 for the vessels on its own 15 test cases: the vessels leave most of the room.
  • Images. The series as delivered, read by GDCM, in the scanner's arbitrary units (maxima from 490 to 9,880 on the test MRs), prone, slices 1 to 1.3 mm apart, pixels 0.63 to 1.13 mm.
  • Start. A mask of the whole image scores a Dice at the median of the test MRs of 0.23 for the fatty tissue, 0.018 for the fibroglandular tissue and 0.001 for the vessels; an Otsu mask 0.41, 0.083 and 0.005. TotalSegmentator has no breast structure on MR.
  • Overlap. DukeBreastMR_Syn takes its cases from the same collection; its pool leaves out the 209 patients with one of the provider's tissue segmentations, these 100 among them, and none of its 60 cases is in this pool. No other dataset of the corpus names this collection.
  • Outside numbers. None: the study's test patients are not named.

No method has a result on every case of this dataset yet.

Case by case

Click a score to look at that result: the image, the truth in green and the method's map in red. Underlined scores have their image; for the others you see the case without the result.

About the data

Test cases
10 test MRs drawn with seed 2026 among the 90 patients outside development, the two scanner makers in proportion, and 10 development MRs. The pool is the 100 patients with 3D segmentations. The study's 70/15/15 split is not published.
Truth
contours by experts 1 the breast outside the fibroglandular tissue and the vessels, 2 the fibroglandular tissue, 3 the vessels: drawn by hand in 3D Slicer on the pre-contrast series by trained annotators under written rules (the breast from the clavicle to the inframammary fold, without the chest wall; the vessels apart from the fibroglandular tissue), every case modified or approved by a breast fellowship-trained radiologist
Licence
CC BY-NC 4.0
10 cases for tuning, never tested

Breast_MRI_320, Breast_MRI_166, Breast_MRI_723, Breast_MRI_337, Breast_MRI_525, Breast_MRI_741, Breast_MRI_238, Breast_MRI_640, Breast_MRI_529, Breast_MRI_636