fidoptBCBM-RadioGenomics, breast cancer brain metastases
Post-contrast T1 MR of the brain, skull-stripped, of a patient treated with Gamma Knife radiosurgery for brain metastases of breast cancer at the University of Minnesota, every metastasis outlined for the treatment (BCBM-RadioGenomics, TCIA).
Read before comparing (7)
- Pool. 165 patients treated with Gamma Knife radiosurgery for brain metastases of breast cancer, 268 sessions; a case is the first session, <patient>-0. 134 patients are in the pool. Held out by the names of their masks: 9 with a resection cavity, 1 with a planning target, 19 with a name that may be an organ at risk or a lesion (a bare 'pons', 'cerebellum', 'vermis', 'thalamus', 'AN'...), and 2 whose sessions start at 1; BCBM-RadioGenomics-169, drawn for development, because one of its masks is a larger region round another.
- Images. The provider's post-contrast T1, skull-stripped with HD-BET and bias-corrected with N4: 0.5 x 0.5 x 1 mm on the drawn cases, values without unit, 0 outside the brain. 1.5 and 3 T scanners. No mask is given.
- Truth. The contours drawn for the treatment, made and reviewed by at least one neurosurgeon and one radiation oncologist, one mask a structure named freely. Every mask but the organs at risk (brainstem, optic nerves and chiasm, cochleae, orbits, lenses, pituitary stalk) is a metastasis, merged into one label. On the drawn cases the masks lie on the image's grid and read 1.1 to 2.8 times brighter than the brain on average, save a cystic cerebellar lesion of BCBM-RadioGenomics-65; all but two of those whose name gives a side lie on that side, the two near the midline. Graded expert.
- Lesions. The test cases hold 1 to 10 metastases, 0.07 to 11 mL in all (median 4.0 mL); 15 of the 41 test lesions are under 0.1 mL, so small lesions weigh on the Dice.
- Start. Over the 13 test cases a whole-image mask scores a Dice of 0.0007 at the median, an Otsu mask 0.0066.
- Overlap. The patients are the University of Minnesota's; no other descriptor names the collection or Minnesota brain data. UCSDBrainMets (UC San Diego), BrainMetsMR and BrainMetsLungMR (Yale) are other hospitals' cohorts.
- Outside numbers. None: the collection has no challenge and no official split.
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
- 13 test patients drawn with seed 2026 among the 121 outside development, then 12 development patients, the last of the same seeded order. A case is a patient's first radiosurgery; one hospital, so no strata. The notes say which patients are held out.
- Truth
- contours by experts every metastasis contoured for the patient's first radiosurgery, by or with a neurosurgeon and a radiation oncologist, merged into one label; the organs at risk drawn beside them are left out
- Source
- www.cancerimagingarchive.net/collection/bcbm-radiogenomics
doi.org/10.7937/rrse-w278
doi.org/10.1038/s41597-025-06066-6
www.cancerimagingarchive.net/wp-content/uploads/BCBM-RadioGenomics-Clinical-data_v01_20250321v2.xlsx
github.com/birra-taha/BreastCancer - Licence
- CC BY 4.0
12 cases for tuning, never tested
BCBM-RadioGenomics-69, BCBM-RadioGenomics-57, BCBM-RadioGenomics-72, BCBM-RadioGenomics-114, BCBM-RadioGenomics-17, BCBM-RadioGenomics-122, BCBM-RadioGenomics-127, BCBM-RadioGenomics-67, BCBM-RadioGenomics-155, BCBM-RadioGenomics-37, BCBM-RadioGenomics-173, BCBM-RadioGenomics-157