Prostate158
Prostate158 expert-annotated biparametric 3T prostate MRI dataset
Biparametric 3 T prostate MRI (T2w, DWI, ADC) of 158 men with suspected prostate cancer from Charité Berlin, with expert segmentations of the central gland, peripheral zone and PI-RADS 4 to 5 lesions, split into 139 training and 19 test cases.
Overview
Prostate158 is a public benchmark for segmenting prostate anatomy and suspected cancer on biparametric MRI. Radiologists at Charité - Universitätsmedizin Berlin assembled it together with collaborators from Radboud University Medical Center, and released it with baseline U-Net models and training code.
Composition
The collection holds 158 examinations of men with suspected prostate cancer: 139 for training and 19 for testing.
Cancer was confirmed by histology in 102 patients (83 training, 19 test); the other 56 served as controls. Every case
has an axial T2-weighted image, a high b-value diffusion image and an ADC map. The data descriptor reports a mean age
of 69 years (SD 9, range 35 to 84), PI-RADS 4 in 55 and PI-RADS 5 in 47 patients, and Gleason grade groups 1 to 5
in 11, 37, 24, 21 and 9 patients. The training archive ships train.csv and valid.csv that split it further.
Acquisition
Patients were scanned between February 2016 and January 2020 on two Siemens 3 T systems (VIDA and Skyra) with a protocol following PI-RADS v2.1. T2w images have 3 mm slices with 0.47 mm in-plane resolution. DWI used b-values up to 1000 s/mm², and the scanner software computed the ADC map. Only the b = 1000 image was kept. All sequences were resampled onto the T2w grid, center-cropped by removing 25% of the margins, N4-corrected (T2w) and stored as NIfTI.
Annotations
Segmentations were drawn in ITK-SNAP. One board-certified radiologist outlined the central gland and peripheral zone on T2w, and a second reviewed them. Both readers independently segmented PI-RADS 4 and 5 lesions on the ADC map, adjusted against T2w and DWI. Reader 1 also segmented lesions on T2w. The test cases carry anatomy and lesion labels from both readers.
Known limitations
- Single center, single vendor and one field strength.
- The data descriptor states an inclusion age over 50, yet reports a minimum age of 35.
- Lesions below PI-RADS 4 are not segmented.
- Resampling and cropping discard the original resolution of the DWI and ADC series.
- The challenge page describes the test set as hidden, while a Zenodo record now offers it for download.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 158 subjects.
Age
mean 69± 9 · range 35 to 84No age bins reported.
Contrast combinations
How many subjects have exactly each set of contrasts.
| T2w | dwi | ADC | Subjects with exactly this set |
|---|---|---|---|
158 |
Contrast / sequence
subjects, values can overlap
- T2-weighted 158 100%
- Diffusion-weighted 158 100%
- ADC map 158 100%
Condition
subjects, values can overlap
- Prostate cancer 102 65%
Scanner vendor
subjects
- Siemens Healthineers 158 100%
Field strength
subjects
- 3 T 158 100%
Country
subjects
- Germany 158 100%
Split
subjects
- Training 139 88%
- Test 19 12%
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
Prostate158 Research Use Agreement
Personal, non-commercial research use only. Each user registers and agrees individually. The data may not be passed on, modified or used to create derivative works, re-identification is forbidden, and the data must not be used for patient care.
The Research Use Agreement on the challenge data page applies to everyone who registers for downloads. The Zenodo records carry no license field, and on 2026-10-09 both zip files downloaded without an account.
What you can do
- No
- Conditional
- No
- Not stated
What you can share
- No
- No
- Not stated
What you must do
- Not stated
- Share alike No
- Yes
- No
- Manuscript review No
- Release code No
- Return results No
- Conditional
Limits
- Yes
- Location limits No
Citation
Adams LC, Makowski MR, Engel G, Rattunde M, Busch F, Asbach P, Niehues SM, Vinayahalingam S, van Ginneken B, Litjens G, Bressem KK. Prostate158 - An expert-annotated 3T MRI dataset and algorithm for prostate cancer detection. Computers in Biology and Medicine 148, 105817 (2022). doi:10.1016/j.compbiomed.2022.105817
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total | 158 | adams2022-dib Specifications Table |
| Studies | total | 158 | adams2022-cbm Abstract |
| Subjects | split=train | 139 | adams2022-dib Characteristics of the Patient Cohort table |
| Subjects | split=test | 19 | adams2022-dib Characteristics of the Patient Cohort table |
| Subjects | condition=prostate_cancer Histologically verified; the remaining 56 patients served as controls | 102 | adams2022-dib Specifications Table |
| Subjects | condition=prostate_cancer;split=train | 83 | adams2022-dib Characteristics of the Patient Cohort table |
| Subjects | condition=prostate_cancer;split=test | 19 | adams2022-dib Characteristics of the Patient Cohort table |
| Subjects | contrast=T2w | 158 | adams2022-cbm Abstract |
| Subjects | contrast=dwi | 158 | adams2022-cbm Abstract |
| Subjects | contrast=ADC | 158 | adams2022-cbm Abstract |
| Subjects | contrast_set=ADC+dwi+T2w All 158 MRIs comprise T2w, DWI and ADC maps | 158 | adams2022-cbm Abstract |
| Subjects | field_strength=3 | 158 | adams2022-dib Section 3.1 |
| Subjects | vendor=siemens Siemens VIDA and Skyra | 158 | adams2022-dib Section 3.1 |
| Subjects | country=DE | 158 | adams2022-dib Specifications Table |
| Studies | split=train "139 MRIs" | 139 | prostate158-repo Accessing the Prostate158 Data Set |
| Studies | split=test "19 additional MRIs" | 19 | prostate158-repo Accessing the Prostate158 Data Set |
| Mean age | total | 69 | adams2022-dib Characteristics of the Patient Cohort table |
| Age SD | total | 9 | adams2022-dib Characteristics of the Patient Cohort table |
| Minimum age | total The inclusion criteria state an age over 50 | 35 | adams2022-dib Characteristics of the Patient Cohort table |
| Maximum age | total | 84 | adams2022-dib Characteristics of the Patient Cohort table |
Sources
The keys used in the table above.
- adams2022-dib Adams et al. 2022, Data in Brief 45, 108739, Dataset of prostate MRI annotated for anatomical zones and cancer paper
- adams2022-cbm Adams et al. 2022, Computers in Biology and Medicine 148, 105817 (abstract) paper
- prostate158-repo kbressem/prostate158 GitHub repository README website
- zenodo-6481141 Prostate158 - Training data, Zenodo record data
- zenodo-6592345 Prostate158 - test data, Zenodo record data