In short
- We only work with de-identified data that we have permission to use.
- Raw patient data never enters the project.
- A dataset being open does not remove the need for ethics and legal review.
- A person always decides whether real surgical material is published.
What data we accept
Open de-identified datasets. Our main source. They were collected with patient consent and ethics committee approval. For reuse, we record a written opinion and the source’s ethics status in each dataset’s card.
New material from partner clinics. Only after approval by the clinic’s ethics committee (IRB) and with informed patient consent. The clinic obtains that approval, not the project.
Data whose de-identification is not confirmed is not accepted. Do not send it through this website or in public issues.
Four layers of de-identification
- Metadata: names, dates, identifiers and other DICOM fields are removed according to a standard profile.
- Vendor-specific fields: everything is removed except a reviewed allow-list.
- Text in the image: every frame is checked with text recognition (OCR) for names, dates and clinic names.
- Indirect identifiers: faces in 3D head scans, rare implants, equipment serial numbers – removed, or the material is rejected.
Automatic checks do not guarantee anonymity, so a person reviews the results too.
If something is wrong
To correct or remove material, follow the privacy process. Report vulnerabilities through the security policy. Details are in the project’s ethics document.