fidopt Interactive view
Brain MR, inter-subject, automatic labels

Calgary-Campinas-359, T1 brain MR between subjects

T1-weighted head MR of two healthy adults scanned on the same scanner model (Siemens, Philips or GE, 1.5 or 3 T), with FastSurfer's anatomical labels on both (CC-359).

17 test pairsbraininter-patientfixed MR, moving MRaligned once by the benchmarkresearch only
Look at the images
Read before comparing (6)
  • Pool. CC-359's T1-weighted volumes of 359 healthy adults aged 29 to 80, about 60 on each of six scanners. The 12 with a manual brain mask are CalgaryCampinas_Seg's and stay out; a scanner's odd subject is unused. Both images of a pair come from the same scanner model.
  • Images. The deposit's volumes byte for byte, whole heads with the face and neck, in each scanner's units, voxels of 0.89 to 1.33 mm. No mask is given.
  • Truth. Graded automatic and ranked apart: FastSurfer v2.5.4's 95 structures (aseg and the DKT cortical regions) on each native volume, drawn or checked by no person, with every structure on all 64 images.
  • Alignment. Each pair starts from an intensity-based elastix rigid transform (rotations up to 18 degrees, 7 at the median), written into the moving image's and its labels' headers. From that start the 17 test pairs have a mean Dice of 0.386 over the 95 labels (0.287 to 0.457) and an HD95 of 5.87 mm (4.59 to 8.46).
  • Overlap. No other descriptor names Calgary, Campinas, CC-359 or CONP; IXI, LUMIR's OpenBHB and every other brain MR set of the corpus come from other cohorts. Of the 64 images, the 56 that share a grid with built stores correlate at 0.82 at most, the highest with CalgaryCampinas_Seg's other subjects of the same scanner (a check that finds PsCBCTThorax_Seg's images in PsCBCTThorax_Syn and correlates a CC-359 volume with its own store at 1.000).
  • Licence. CC BY-ND 4.0: the cases are never uploaded, the private copy included. The starting transforms are computed from these images, so they stay on the reference machine (alignment/ is ignored by git); prepare.py pins each one's sha256 and puts it back for a build there.

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

Pair by pair

Click a score to look at that result: the fixed image, and over it the moving image after registration, with the annotations. Underlined scores have their image; for the others you see the pair without the result. A green cell gained against no registration on that pair, a red one lost.

About the data

Test pairs
The 347 subjects without a manual mask, paired at random within each of the six scanners into 173 pairs (prepare.py --pool). 17 test pairs drawn with seed 2026, the scanners in proportion, and 15 development pairs. The deposit has no split.
Truth
contours by a program FastSurfer v2.5.4 FastSurferCNN aparc.DKTatlas+aseg.deep, run by the corpus on each native whole-head T1 volume: subcortical structures and the DKT cortical parcellation, brought back to the image's grid by nearest neighbour. No person drew or checked them.
Licence
CC BY-ND 4.0 (the dataset's download page; CONP's DATS.json: CC BY-ND)
15 pairs for tuning, never tested

CC0009__CC0033, CC0074__CC0094, CC0163__CC0154, CC0224__CC0226, CC0256__CC0270, CC0303__CC0301, CC0058__CC0016, CC0068__CC0089, CC0160__CC0153, CC0231__CC0192, CC0277__CC0259, CC0338__CC0329, CC0045__CC0059, CC0076__CC0104, CC0146__CC0127