Medical Segmentation Decathlon
Medical Segmentation Decathlon (MSD)
Ten 3D segmentation tasks in one benchmark: brain tumours, heart, liver, hippocampus, prostate, lung, pancreas, hepatic vessels, spleen and colon, with 2,633 CT and MRI volumes in NIfTI under CC BY-SA 4.0.
Overview
The Medical Segmentation Decathlon is a benchmark of ten 3D segmentation tasks, first run as a challenge at MICCAI 2018 to test whether one method can solve very different problems without manual tuning per task. It remains a common test bed for general-purpose segmentation methods.
Composition
| Task | Target | Modality | Volumes (train + test) |
|---|---|---|---|
| 01 Brain tumour | Oedema, enhancing and non-enhancing tumour | MRI: T1w, T1w post-contrast, T2w, FLAIR | 484 + 266 |
| 02 Heart | Left atrium | MRI | 20 + 10 |
| 03 Liver | Liver and tumour | Contrast CT | 131 + 70 |
| 04 Hippocampus | Hippocampus head and body | T1w MRI | 263 + 131 |
| 05 Prostate | Central gland and peripheral zone | MRI: T2w, ADC | 32 + 16 |
| 06 Lung | Lung tumour | CT | 64 + 32 |
| 07 Pancreas | Pancreas and mass | Portal venous CT | 282 + 139 |
| 08 Hepatic vessel | Vessels and tumour | Portal venous CT | 303 + 140 |
| 09 Spleen | Spleen | Portal venous CT | 41 + 20 |
| 10 Colon | Colon cancer primaries | Portal venous CT | 126 + 64 |
Each task is a TAR file with a JSON descriptor, labelled training images and unlabelled test images, all converted to NIfTI in a common RAS orientation.
Acquisition
Sources include BraTS 2016/2017, the Left Atrial Segmentation Challenge (King's College London), LiTS, Vanderbilt University Medical Center, Radboud University Medical Center, Stanford lung cancer data on TCIA, and Memorial Sloan Kettering Cancer Center (pancreas, hepatic vessel, spleen, colon). Protocols vary within and across tasks.
Annotations
Experts drew or corrected labels, manually (e.g. hippocampus, colon) or semi-automatically with manual correction (hepatic vessels, spleen). The hippocampus cohort has 90 healthy adults and 105 with a psychotic disorder.
Known limitations
- Subject counts and demographics are missing for most tasks; volumes are not patients.
- The website lists 282 + 139 pancreas volumes but 420 in total, as does the paper.
- Brain tumour data overlap with BraTS releases.
Cohort
Aggregate numbers from the sources below. Bars are relative to the largest value.
Modality
scans
- CT 1,411 54%
- MRI 1,222 46%
Condition
subjects, values can overlap
- Psychotic disorder 105
- Healthy control 90
- Schizophrenia 56
Anatomy
scans
- Brain 750 28%
- Liver 644 24%
- Pancreas 420 16%
- Hippocampal formation 394 15%
- Colon 190 7%
- Lung 96 4%
- Spleen 61 2%
- Prostate 48 2%
- Heart 30 1%
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
Download from AWS or Google Drive; one TAR file per task. Test labels are not released.
Creative Commons Attribution-ShareAlike 4.0 International
Use, share and adapt for any purpose with credit. Anything you share that is built on the data must use the same license.
What you can do
- Yes
- Yes
- Yes
- Yes
What you can share
- Yes
- Conditional
- Conditional
What you must do
- Yes
- Yes
- No
- No
- Manuscript review No
- Release code No
- Return results No
- Delete after use No
Limits
- No
- Location limits No
Citation
Simpson AL, Antonelli M, Bakas S, et al. A large annotated medical image dataset for the development and evaluation of segmentation algorithms. arXiv:1902.09063 (2019). Antonelli M, Reinke A, Bakas S, et al. The Medical Segmentation Decathlon. Nature Communications 13, 4128 (2022).
All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Scans | total Sum of the N column for the ten tasks (3D or 4D volumes, training and test) | 2,633 | simpson2019 Table 1 |
| Scans | modality=MR Sum of N column: Task01 brain (750), Task04 hippocampus (394), Task05 prostate (48), Task02 heart (30) | 1,222 | simpson2019 Table 1 |
| Scans | modality=CT Sum of N column: Task03 liver (201), Task08 hepatic vessels (443), Task07 pancreas (420), Task10 colon (190), Task06 lung (96), Task09 spleen (61) | 1,411 | simpson2019 Table 1 |
| Scans | anatomy=brain Task01 BrainTumour; website split: 484 training + 266 test; multimodal MRI from BraTS 2016/2017 | 750 | simpson2019 Table 1 |
| Scans | anatomy=hippocampus Task04 Hippocampus; website split: 263 training + 131 test; T1w MPRAGE | 394 | simpson2019 Table 1 |
| Scans | anatomy=prostate Task05 Prostate; website split: 32 training + 16 test; T2w and ADC | 48 | simpson2019 Table 1 |
| Scans | anatomy=heart Task02 Heart (left atrium); website split: 20 training + 10 test | 30 | simpson2019 Table 1 |
| Scans | anatomy=liver Task03 Liver (201) + Task08 HepaticVessel (443); website splits 131 + 70 and 303 + 140 | 644 | simpson2019 Table 1 |
| Scans | anatomy=pancreas Task07 Pancreas; the website gives 282 training + 139 test | 420 | simpson2019 Table 1 |
| Scans | anatomy=colon Task10 Colon; website split: 126 training + 64 test | 190 | simpson2019 Table 1 |
| Scans | anatomy=lung Task06 Lung; website split: 64 training + 32 test | 96 | simpson2019 Table 1 |
| Scans | anatomy=spleen Task09 Spleen; website split: 41 training + 20 test | 61 | simpson2019 Table 1 |
| Scans | condition=glioma Glioblastoma or lower-grade glioma | 750 | simpson2019 p. 3 (Task01_BrainTumour) |
| Scans | condition=liver_cancer Primary and metastatic liver tumours (Task03 and Task08) | 644 | simpson2019 pp. 4 and 6 (Task03_Liver, Task08_HepaticVessel) |
| Scans | condition=pancreatic_neoplasm IPMN, neuroendocrine tumours or PDAC | 420 | simpson2019 pp. 5-6 (Task07_Pancreas) |
| Scans | condition=colorectal_cancer Primary colon cancer | 190 | simpson2019 p. 6 (Task10_Colon) |
| Scans | condition=nsclc | 96 | simpson2019 p. 5 (Task06_Lung) |
| Subjects | condition=healthy Task04 Hippocampus source cohort (195 labelled atlases) | 90 | simpson2019 p. 5 (Task04_Hippocampus) |
| Subjects | condition=psychotic_disorder Task04: 56 schizophrenia, 32 schizoaffective, 17 schizophreniform | 105 | simpson2019 p. 5 (Task04_Hippocampus) |
| Subjects | condition=schizophrenia | 56 | simpson2019 p. 5 (Task04_Hippocampus) |
Sources
The keys used in the table above.
- simpson2019 Simpson et al. 2019, dataset paper, arXiv 1902.09063 paper
- msd-website Medical Segmentation Decathlon website (task list) website
- antonelli2022 Antonelli et al. 2022, The Medical Segmentation Decathlon, Nature Communications paper