SKM-TEA
Stanford Knee MRI with Multi-Task Evaluation
Raw multi-coil k-space and scanner DICOM images of 155 quantitative double-echo steady-state (qDESS) knee MRI scans from Stanford, with manual cartilage and meniscus masks and pathology bounding boxes for reconstruction, segmentation and detection benchmarks.
Overview
SKM-TEA (Stanford Knee MRI with Multi-Task Evaluation) is a knee MRI collection from Stanford University, released through the Stanford AIMI Center and presented at the NeurIPS 2021 Datasets and Benchmarks track. It pairs raw multi-coil k-space with the scanner's own DICOM images and dense labels, so one set of scans can test accelerated reconstruction, tissue segmentation, pathology detection and the cartilage and meniscus T2 values derived from them.
Composition
There are 155 patients, each with one quantitative double-echo steady-state (qDESS) knee scan. Each scan holds two co-registered 3D echoes, from which T2 relaxation maps can be computed. The release fixes a split of 86 training, 33 validation and 36 test scans. The test set holds the patients who also had arthroscopic surgery; the remaining scans were split at random. Data come as HDF5 (k-space, coil sensitivity maps, SENSE reconstructions, image arrays), DICOM and NIfTI masks. The documentation gives the size as about 900 GB compressed and 1.6 TB unpacked.
Acquisition
All scans were acquired at Stanford Healthcare on two 3 T GE MR750 systems with 2x1 parallel imaging and elliptical sampling. Missing k-space lines were filled in with GE's ARC method, and the result is treated as fully sampled. The paper reports 15 or 16 receive coils and 80 to 88 slices per scan, at an in-plane resolution of 0.38 x 0.31 mm.
Annotations
Two researchers with 3 to 4 years of knee MRI experience, supervised by two musculoskeletal radiologists, outlined patellar, femoral, medial and lateral tibial cartilage and the medial and lateral meniscus slice by slice on the DICOM images. A second mask set was registered to the SENSE reconstructions to undo the scanner's gradient warping. The same annotators drew 3D bounding boxes for 16 pathology categories (meniscal tears, ligament and cartilage lesions, joint effusion) taken from the radiology reports. Each scan was labelled by one person.
Known limitations
- Single center, one vendor and one sequence.
- Coil counts differ between scans, and the AIMI page states 8 or 15 coils where the paper states 15 or 16.
- Scanner DICOM images and T2 maps carry vendor post-processing and are not meant as reconstruction targets.
- No demographics or per-pathology counts are given in the paper or on the dataset pages.
- The research use agreement limits use to personal, non-commercial research and forbids redistribution and derivative works.
Cohort
Aggregate numbers from the sources below. Bars are relative to the 155 subjects.
Anatomy
subjects
- Knee 155 100%
Scanner vendor
subjects
- GE HealthCare 155 100%
Field strength
subjects
- 3 T 155 100%
Split
scans
- Training 86 55%
- Test 36 23%
- Validation 33 21%
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
Create an account and accept terms online
Sign in and accept the Stanford research use agreement. The AIMI page links to Redivis for the download; the older Stanford AIMI shared-datasets portal still lists the dataset.
Stanford University School of Medicine Dataset Research Use Agreement (AIMI)
Free for personal, non-commercial research only. You may not share the data or the download link, may not create derivative works, and must not try to re-identify patients. Stanford sells a separate commercial license.
The portal shows the Stanford University School of Medicine Research Use Agreement with "SKM-TEA" as the dataset name. The paper (Appendix A.5) also says the public data are distributed under the Stanford School of Medicine terms. The portal's metadata API names the Open Use of Data Agreement v1.0, but the agreement shown to users is the research use agreement. The code repository is MIT licensed; that license does not cover the data.
What you can do
- No
- Not stated
- No
- Not stated
What you can share
- No
- No
- Share trained models Not stated
What you must do
- Yes
- Share alike No
- Yes
- Ethics approval No
- Manuscript review No
- Release code No
- Return results No
- No
Limits
- Yes
- Location limits No
Commercial license: Yes Companies can apply for a one-year license per dataset, reviewed by a Stanford committee. The page lists an annual fee of USD 70,000 per dataset for agreements contracted in FY25. Details
Citation
Desai AD, Schmidt AM, Rubin EB, Sandino CM, Black MS, Mazzoli V, Stevens KJ, Boutin R, Ré C, Gold GE, Hargreaves BA, Chaudhari AS. SKM-TEA: A Dataset for Accelerated MRI Reconstruction with Dense Image Labels for Quantitative Clinical Evaluation. NeurIPS Datasets and Benchmarks Track (2021). arXiv:2203.06823. Dataset DOI: https://doi.org/10.71718/2ghb-nv62
@inproceedings{desai2021skm,
title={SKM-TEA: A Dataset for Accelerated MRI Reconstruction with Dense Image Labels for Quantitative Clinical Evaluation},
author={Desai, Arjun D and Schmidt, Andrew M and Rubin, Elka B and Sandino, Christopher Michael and Black, Marianne Susan and Mazzoli, Valentina and Stevens, Kathryn J and Boutin, Robert and Re, Christopher and Gold, Garry E and others},
booktitle={Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2)},
year={2021}
} All numbers
Every number on this page, as stored in stats.csv, with its source.
| Measure | Breakdown | Value | Source |
|---|---|---|---|
| Subjects | total patients at Stanford Healthcare | 155 | desai2021 Section 4.1 |
| Scans | total one qDESS knee scan per patient | 155 | aimi-skm-tea |
| Scans | split=train | 86 | aimi-skm-tea |
| Scans | split=validation | 33 | aimi-skm-tea |
| Scans | split=test patients who also had arthroscopic surgery (desai2021 Section 4.2) | 36 | aimi-skm-tea |
| Subjects | anatomy=knee | 155 | desai2021 Section 4.1 |
| Subjects | field_strength=3 two 3 T GE MR750 scanners | 155 | desai2021 Section 4.1 |
| Subjects | vendor=ge | 155 | desai2021 Section 4.1 |
Sources
The keys used in the table above.
- desai2021 Desai et al., SKM-TEA - A Dataset for Accelerated MRI Reconstruction with Dense Image Labels for Quantitative Clinical Evaluation (arXiv:2203.06823v1) paper
- aimi-skm-tea Stanford AIMI SKM-TEA Knee MRI dataset page website
- github-skm-tea-dataset SKM-TEA dataset documentation (DATASET.md in the StanfordMIMI/skm-tea repository) website