PROMISE12
Prostate MR Image Segmentation 2012 (MICCAI Grand Challenge)
100 transverse T2-weighted prostate MRI scans from four centres in Norway, the US, the UK and the Netherlands, with reference segmentations of the prostate, from the PROMISE12 segmentation challenge held at MICCAI 2012.
Overview
PROMISE12 is the data of the Prostate MR Image Segmentation challenge, run at MICCAI 2012 in Nice by groups from Radboud University Nijmegen Medical Centre, Rutgers University and University College London. It was built to compare interactive, semi-automatic and automatic methods for delineating the whole prostate on MRI across centres, scanner vendors and protocols, and it has since served as a common benchmark for prostate segmentation.
Composition
The set holds 100 transverse T2-weighted MR scans of the prostate, 25 from each of four centres. The organisers split them at random, stratified by centre, into 50 training cases, 30 test cases and 20 cases for the live challenge held at the workshop. Only the training masks were released during the challenge. Since 2023 all three parts, each case with its reference mask, can be downloaded from Zenodo as MetaImage (MHD/RAW) files.
Acquisition
The scans come from Haukeland University Hospital (Norway), Beth Israel Deaconess Medical Center (USA), University College London (UK) and Radboud University Nijmegen Medical Centre (Netherlands). Field strength was 1.5 T at Haukeland, 3 T at Beth Israel and Radboud, and both at UCL. Beth Israel used GE scanners, the other centres Siemens. The two centres with an endorectal coil are Haukeland and Beth Israel. In-plane resolution ranges from 0.25 to 0.75 mm and slice thickness from 2.2 to 4 mm. The scans were made for prostate cancer detection or staging.
Annotations
Each centre supplied a slice-by-slice outline of the prostate capsule drawn by an experienced reader in 3D Slicer or MeVisLab. One expert who had read more than 1000 prostate MRIs then checked all outlines for consistency and corrected them where needed; these form the reference standard. For the test and live challenge cases, a second, less experienced observer segmented the prostate blind to the reference, and the challenge score is scaled against that observer.
Known limitations
- Small cohort, one sequence, and a single label for the whole gland; zones and lesions are not marked.
- Clinical stage, cancer presence and lesion location are unknown to the organisers, and no age or other patient data are included. The challenge site states that the cases include both benign disease, such as benign prostatic hyperplasia, and prostate cancer.
- Coil use, resolution and contrast differ strongly between centres.
- The second observer's segmentations are not part of the download.
Cohort
Aggregate numbers from the sources below. Bars are relative to the largest value.
Contrast / sequence
scans, values can overlap
- T2-weighted 100 100%
Scanner vendor
scans
- Siemens Healthineers 75 75%
- GE HealthCare 25 25%
Country
scans
- Norway 25 25%
- United States 25 25%
- United Kingdom 25 25%
- Netherlands 25 25%
Split
scans
- Training 50 50%
- Test 50 50%
License and access
Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer
Download without an account
PROMISE12 data terms (LICENSE.TXT on Zenodo)
A two-sentence notice that permits research use and requires citing the PROMISE12 overview paper by Litjens et al. (Medical Image Analysis 2014). It says nothing about commercial use, redistribution or derived data. Zenodo lists the license as "Other (Attribution)".
What you can do
- Not stated
- Not stated
- Create derived data Not stated
- Conditional
What you can share
- Share the data Not stated
- Share derived data Not stated
- Share trained models Not stated
What you must do
- Yes
- Share alike No
- Sign an agreement No
- Ethics approval No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Citation
Litjens G, Toth R, van de Ven W, Hoeks C, Kerkstra S, van Ginneken B, et al. Evaluation of prostate segmentation algorithms for MRI: the PROMISE12 challenge. Medical Image Analysis 18(2), 359-373 (2014). doi:10.1016/j.media.2013.12.002
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Scans | total Transverse T2-weighted MR images, 25 per centre | 100 | litjens2014 Section 2.1 |
| Scans | contrast=T2w | 100 | litjens2014 Section 2.1 |
| Scans | split=train Released with reference segmentations | 50 | litjens2014 Section 2.1 |
| Scans | split=test Sum of the 30 online test cases and the 20 live challenge cases, two disjoint sets; reference segmentations for both are now public | 50 | litjens2014 Section 2.1 |
| Scans | country=NO Haukeland University Hospital; 1.5 T with endorectal coil | 25 | litjens2014 Table 1 |
| Scans | country=US Beth Israel Deaconess Medical Center; 3 T with endorectal coil | 25 | litjens2014 Table 1 |
| Scans | country=GB University College London; 1.5 T and 3 T, no endorectal coil | 25 | litjens2014 Table 1 |
| Scans | country=NL Radboud University Nijmegen Medical Centre; 3 T, no endorectal coil | 25 | litjens2014 Table 1 |
| Scans | vendor=siemens Sum of HK, UCL and RUNMC, 25 each | 75 | litjens2014 Table 1 |
| Scans | vendor=ge BIDMC | 25 | litjens2014 Table 1 |
Sources
The keys used in the table above.
- litjens2014 Litjens et al. 2014, Medical Image Analysis (author manuscript, PMC4137968) paper
- promise12-details PROMISE12 challenge website, Details page website
- zenodo-8026660 PROMISE12 data on Zenodo, record 8026660 (training_data.zip, test_data.zip, livechallenge_test_data.zip) data