fidopt Interactive view
Brain MR, segmentation

Calgary-Campinas-359, manual brain masks

One T1-weighted head MR of a healthy adult, from a Siemens, Philips or GE scanner at 1.5 or 3 T, with the brain mask an expert drew by hand (CC-359).

10 test casesbrainimage MRresearch only
Look at the images
Read before comparing (6)
  • Pool. CC-359 holds T1-weighted volumes of 359 healthy adults aged 29 to 80, about 60 on each of six scanners (Siemens, Philips and GE at 1.5 and 3 T). Only the 12 subjects with a manual mask are here; CalgaryCampinas registers the other 347.
  • Truth. One expert's manual mask: brain, cerebellum, brainstem and ventricles. The deposit's second manual mask, BEaST's mask corrected by hand, agrees with it at a Dice of 0.965 at the median of the test cases (0.948 to 0.975): the ceiling of a method.
  • Images. The deposit's volumes byte for byte, whole heads with the face, in each scanner's own units (Philips as 32-bit floats, the others 16-bit integers); voxels of 0.89 to 1.33 mm. No mask is given.
  • Start. Labelling the whole image as brain scores a Dice of 0.196 at the median of the test cases (0.128 to 0.245), an Otsu mask 0.550 (0.479 to 0.632). TotalSegmentator 2.18.0's total_mr brain scores 0.919 on CC0003, 0.941 on CC0180 and 0.932 on CC0240, its brain 10 to 17% larger than the expert's.
  • Overlap. No other descriptor names Calgary, Campinas, CC-359 or CONP. No built image of the corpus shares the voxels of these 12: 10 have no other store on their grid, and the two on a common 192 x 256 x 256 grid correlate at 0.65 at most with its 65 stores (a check that finds PsCBCTThorax_Seg's images in PsCBCTThorax_Syn, and correlates CC0003 with its own store at 1.000).
  • Licence. CC BY-ND 4.0: the cases are never uploaded, the private copy included, and nothing computed from the images is shared.

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
The 12 subjects with a manual mask, two of each vendor and field strength: 10 test cases drawn with seed 2026, the three vendors in proportion, and the 2 others for development. The deposit has no split.
Truth
contours by experts the deposit's manual brain segmentation (Manual/<subject>_manual.nii.gz), drawn by an expert on twelve randomly selected subjects (Souza et al. 2018)
Licence
CC BY-ND 4.0 (the dataset's download page; CONP's DATS.json: CC BY-ND)
2 cases for tuning, never tested

CC0181, CC0300