fidoptMSD Prostate
Multiparametric prostate MR from Radboud: the transverse T2 and the ADC map as one two-channel image, with the peripheral zone and the rest of the gland segmented (Medical Segmentation Decathlon, Task05).
10 test casesprostateimage MR
Read before comparing (8)
- Image. The archive's 4D NIfTI written as one image of two components a voxel, the T2 then the ADC, values untouched. The provider had resampled the ADC onto the T2's grid: it is visibly blurrier, and zero outside the body on some cases. Neither carries a unit: the T2 is on the scanner's arbitrary scale, the ADC on that of 10^-6 mm^2/s.
- Labels. The archive says 1 PZ and 2 TZ, but label 2 is the gland outside its peripheral zone, the source study's central gland: together they cover the gland, the peripheral zone a posterior crescent, as in ProstateISBI2013 and PROSTATEx_Seg.
- Truth. Drawn by hand on the T2 for a zonal segmentation study (Litjens et al., MICCAI 2012), which used three observers; the archive does not say whose outline it holds. One reader a case, no second reading published for the decathlon.
- Checks. On all 20 cases each map holds 0, 1 and 2 on its image's grid, and the peripheral zone lies behind and below the rest. prostate_16 and prostate_38 are much enlarged glands (304 and 190 mL outside the peripheral zone), the outline following them.
- TotalSegmentator. Version 2.18 (total_mr) on prostate_06 overlaps the union of the two labels at a Dice of 0.79, but has no zones: a method that writes only the gland scores 0 on both labels.
- Tolerance. The challenge's clinicians took 4 mm as the error they would accept on this task's surfaces: the surface Dice here is at 4 mm.
- Overlap. PROMISE12's Radboud images are decathlon images, voxel for voxel; the seven that are training cases are left out. The 20 built images match none of the corpus's other prostate MR (ProstateISBI2013, PROSTATEx_Seg, QINProstateRepeat), by size and at a correlation of 0.6 at most. PROMISE12's 50 training cases were not compared.
- Outside numbers. None: the decathlon ranks its entries on its hidden test set. The test cases are public decathlon training cases with their labels, so a network trained on the task has seen them.
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
The first test caseClick to look at every case.
- Test cases
- 10 test cases drawn with seed 2026 and 10 development cases from 24 training cases: the decathlon's 32 less the seven that are PROMISE12 images (prostate_20, 21, 24, 25, 32, 35, 37) and prostate_18, whose label is wrong (prepare.py). The labels of the decathlon's 16 test cases were never released.
- Truth
- contours by experts the decathlon's two labels, drawn by hand on the T2 at Radboud for a zonal segmentation study (Litjens et al., MICCAI 2012): 1 the peripheral zone (PZ) and 2 what the archive calls the transition zone (TZ), in fact the gland outside its peripheral zone, that study's central gland; who drew them is not published
- Source
- medicaldecathlon.com
arxiv.org/abs/1902.09063
doi.org/10.1038/s41467-022-30695-9
msd-for-monai.s3-us-west-2.amazonaws.com/Task05_Prostate.tar
pubmed.ncbi.nlm.nih.gov/23286075
zenodo.org/records/8026660 - Licence
- CC BY-SA 4.0
10 cases for tuning, never tested
prostate_39, prostate_00, prostate_41, prostate_31, prostate_01, prostate_02, prostate_04, prostate_46, prostate_43, prostate_28