fidoptSELMA3D cell nuclei of the human brain, light-sheet microscopy
A 200 x 200 x 200 voxel patch of a light-sheet volume of cleared human brain tissue, its cell nuclei drawn by experts (SELMA3D 2024).
10 test caseshuman brainimage FLUORESCENCE
Read before comparing (6)
- Truth. One binary mask a patch, the cell nuclei drawn by hand; separate cell nuclei that touch are one region, and no instance is scored.
- Ranked apart. Tiles of human brain tissue, from one donor or a few (the deposit's sample names a donor, its protocol brains of human males): the test patches are separate tiles, none touching another, but not independent subjects.
- Units. The headers give no voxel size, so distances are in voxels and the surface tolerance is one voxel. The deposit gives the acquisition at 0.54 x 0.54 x 5 µm without saying whether the patches keep it.
- Contact. Across the split the best face correlation is 0.38 and the best overlap of 1 to 15 slices 0.14, where patch_007 and patch_008 score 0.93; no two test patches touch (best face correlation 0.28).
- Start. The whole patch named nucleus scores a median Dice of 0.04 on the test patches, its bright voxels (Otsu) 0.42 (0.29 to 0.71).
- Public. These are SELMA3D's public annotated training patches: a network trained on them has seen the test cases.
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 deposit's gzip headers keep each patch's tile number (1485 to 11083). Tiles 5661 and 5662 (patch_007 and 008) are neighbours that overlap by about five slices: they are the 2 development patches, and the 10 others the test patches.
- Truth
- contours by experts the deposit's pixel-wise masks of the cell nuclei, drawn in 3D in virtual reality by four expert annotators, then verified by an expert and reviewed by two organisers; the corpus makes every nonzero value 1
- Source
- www.ebi.ac.uk/biostudies/BioImages/studies/S-BIAD1196
doi.org/10.6019/S-BIAD1196
selma3d.grand-challenge.org/data
doi.org/10.1016/j.cell.2020.01.030 - Licence
- CC BY 4.0 (BioImage Archive deposit S-BIAD1196)
2 cases for tuning, never tested
patch_007, patch_008