These terms cover your use of the OpenMSK web service at openmsk.com. By uploading data, you agree to them. OpenMSK is an independent academic project, not a university or hospital service.
OpenMSK is also free and open-source. You can download it from github.com/gattia/OpenMSK, install it on your own computer, and run the same analyses there, in which case none of this applies and no data ever leaves your machine. The repository has the code, the model weights, and install instructions. This website exists to make the pipeline easy to try without setting any of that up.
OpenMSK is for non-commercial research and education. It isn't a medical device, and hasn't been reviewed or cleared by the FDA or any other regulator. Don't use it for clinical diagnosis, treatment planning, or any decision about a patient. Results are generated automatically, aren't reviewed by a clinician, and may be wrong.
Segmentation and shape fitting use neural networks, but they only produce measurements — nothing here makes a decision about anyone. BScore and all other outputs are research numbers, not validated clinical measures.
When you upload data, you're confirming that:
If you realise you've uploaded something you shouldn't have, email us and we'll delete it. Please also don't resell access, submit enough jobs to degrade the service for others, or poke at other people's jobs. We may turn down any submission.
Uploads are processed on our server, hosted on commercial cloud infrastructure in the United States, so if you're outside the US your data is transferred here. Uploads and downloads use HTTPS, and the disks are encrypted at rest.
Once your image is read, it's converted to NRRD — a format that holds voxel data and geometry only, and can't carry DICOM metadata. Converting therefore strips the DICOM headers, and the image is renamed so your original filename isn't kept either. Where the analysis needs more than one echo (qDESS T2 mapping uses two), each echo is stored separately so the results stay reproducible.
From the headers we keep a fixed list of scan acquisition parameters, and discard everything else:
Kept: echo times, repetition time, inversion time, flip angle,
echo train length, number of averages, pixel bandwidth, sequence name and variant,
field strength, scanner manufacturer and model, receive coil, slice thickness,
voxel spacing, image dimensions, patient position and orientation, body part,
laterality, sex, age in years (90 and over recorded as 90), and height and weight
where the headers carry them. Also two vendor private fields encoding gradient
spoiling, (0019,10B6) and (0019,10B7), which are needed
to compute T2 from qDESS and describe the pulse sequence, not the person.
Discarded: everything else — patient name, patient and accession numbers, all dates, institution, referring physician, device serial numbers, and all other private tags. Study and series identifiers are kept only as a one-way hash.
Subject information you enter yourself. The upload form also asks, optionally, for the subject's sex, age, height and weight. These are the covariates almost any musculoskeletal analysis needs, and most uploads carry none of them: a NIfTI or NRRD file has no headers at all, and research DICOMs often have these fields blanked by de-identification. Every one of them is optional. Leaving them all blank changes nothing about your results, and we don't ask twice.
What you enter is kept alongside the acquisition parameters above, on the same terms as everything else here, and may be published in de-identified form (§5). It is stored separately from anything read from the headers, and the two never overwrite each other — where they disagree, that is itself worth knowing. Age is recorded in whole years and capped the same way, so 90 means "90 or older". Sex, age under 90, height and weight are not among HIPAA Safe Harbor's eighteen identifiers, so recording them stays inside the de-identification standard §3 asks you to meet. Don't enter anything else about the subject, here or in a filename.
The file you uploaded, and anything extracted from it, is deleted when your job finishes, or within 24 hours, whichever comes first.
Your results come as a zip at a download link, which we also email you. That link isn't password-protected, so anyone who has it can download your results — don't forward it. The zip is deleted 24 hours after your job finishes.
We keep the converted image, the derived results, the acquisition parameters above, and any subject information you chose to enter, and use them to improve OpenMSK and for musculoskeletal imaging research. Keeping them is a condition of using the free hosted service. If you'd prefer to keep your data entirely in-house, you can install OpenMSK from GitHub and run the same pipeline on your own computer instead (§1).
You agree that what we keep may go into our training and evaluation data, be shared with other researchers, and be published in de-identified form — we periodically release OpenMSK training and benchmark data so results can be reproduced. Anything published contains only converted images, derived results, acquisition parameters, and subject information you entered; never original files, filenames, or email addresses. Published data can't be recalled. We don't sell any of it.
We also store a SHA-256 checksum of the image data, so we can spot the same scan being uploaded twice and avoid double-counting it. A checksum is one-way — it can't reproduce or reveal the image.
We use your email to send you your results, and occasionally to contact you about OpenMSK — major changes, related papers, or collaboration. We keep it indefinitely, don't sell it, and will stop emailing you or delete it if you ask.
To have your data deleted, email info@openmsk.com with your job ID (it's on the results page and in your notification email) or the email address you used. That includes the derived results. One person handles these, so allow up to 90 days. We may keep the duplicate-detection checksum, which identifies no one, and we can't pull back anything already published.
OpenMSK builds on components licensed for non-commercial academic use, and the neural shape modelling method it uses is the subject of a pending patent application filed by Stanford University. Nothing in these terms grants you any licence — patent, copyright, or otherwise — beyond using this service for non-commercial research. If you want to use OpenMSK or its methods commercially, get in touch first.
The example scan and results we publish on the home page are a maintainer's own knee, shared deliberately. You're welcome to use them for research and teaching, including in publications, provided you cite the paper below. They are not for clinical use, and the same "research use only" terms in §2 apply to them.
Results we produce for you are yours to use in your research and to publish. If OpenMSK contributes to your work, please cite it:
Goyal A, Belibi F, Sahani V, Pedersen R, Vainberg Y, Williams A, Chu C, Haddock B, Gold G, Chaudhari A, Kogan F, Gatti A. Automating Imaging Biomarker Analysis for Knee Osteoarthritis Using an Open-Source MRI-Based Deep Learning Pipeline. medRxiv 2025. doi:10.1101/2025.02.21.25322094
Guidance for investigators, not legal advice.
If your board needs something this doesn't cover, get in touch.
OpenMSK runs on one server, maintained as best we can. It may be slow or unavailable without warning, jobs may be lost, and we may change or shut it down at any time. Don't treat it as the only copy of anything, and download your results before they expire.
THE SERVICE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, ACCURACY, AND NON-INFRINGEMENT. TO THE FULLEST EXTENT PERMITTED BY LAW, WE ARE NOT LIABLE FOR ANY DAMAGES OR LOSSES ARISING FROM YOUR USE OF THE SERVICE, INCLUDING LOST DATA OR LOST RESEARCH TIME.
If data you upload in breach of §3 causes a claim against us, you agree to cover it.
We may update these terms; the version and date above will change, and whichever version is current governs uploads made while it's in effect. A copy of the version that applied to your job is included in your results zip, so you always keep the terms you actually agreed to. We won't retroactively change how we use data already uploaded under an earlier version. If you need a superseded version, ask us.
California law governs these terms, and any dispute goes to the courts of California. Questions: info@openmsk.com.