De-identification for research and AI
Your archive is a research asset. Release it without releasing your patients.
We de-identify DICOM imaging at archive scale: headers, private tags, UIDs and burned-in text. Studies stay linked and dates stay meaningful, so researchers and AI teams get data they can use, and your patients stay anonymous.
Where identifiers hide
Removing the patient name is the easy part. The risk is in the places people forget.
Standard DICOM tags
Names, IDs, dates, institutions and physicians, handled with rules based on the DICOM confidentiality profiles.
Private and vendor tags
Proprietary tags that copy patient data where no standard profile looks. Every tag is accounted for, not just the known ones.
Burned-in text
Names and IDs drawn into ultrasound frames, screenshots and scanned documents. Detected and masked in the pixels.
Study and image UIDs
Instance UIDs carry clues back to the source. They are replaced consistently, so links between studies, series and images stay intact.
Data researchers can actually use
Patients stay linked
Identifiers replaced with consistent pseudonyms, so all of a patient's studies share one research ID.
Time stays real
Dates shifted by a consistent offset per patient. Intervals between studies survive; real dates don't.
You choose what stays
Age, sex, modality, body part and other fields kept or generalized according to the project's needs.
Rules per source
Every institution and vendor stores data differently. Rule sets are tuned to each source, not one template for all.
How a project runs
Designed with you, run in your environment, documented for your privacy office.
- DesignDefine the dataset, which fields stay, how patients are linked and where the output goes.
- AssessScan a sample of the source to find the private tags and burned-in text this archive actually contains.
- De-identifyRun at archive scale on servers inside your network.
- VerifyCheck the output against the rules, with sampled review of headers and pixels.
- DeliverHand over the dataset and a report documenting every rule applied, for your privacy office or expert.
Built around HIPAA's two methods
HIPAA allows de-identification by Safe Harbor, removing 18 specified identifiers, or by Expert Determination, where a qualified expert certifies the risk of re-identification is very small. Our rules can remove the Safe Harbor identifiers, and our reports give your expert what they need.
Whether a dataset is de-identified under HIPAA is your organization's determination. We make it a well-documented one.
Common questions
What is the difference between de-identification and anonymization?
People use the terms loosely. In practice, anonymization usually means identifiers are removed with no way back, while de-identification may replace them with consistent codes (pseudonymization) so studies from the same patient stay linked. We support both and agree the approach with you during design.
Which HIPAA de-identification method does this support?
HIPAA recognizes two methods: Safe Harbor, which removes 18 specified identifiers, and Expert Determination, where a qualified expert certifies the re-identification risk is very small. Our rules can be configured to remove the Safe Harbor identifiers. Whether a dataset meets either standard is determined by your organization, and for Expert Determination by your expert; we provide the documentation they need.
Can patients stay linked across studies?
Yes. Identifiers can be replaced with consistent pseudonyms, so every study from the same patient carries the same research ID without revealing who they are.
What about dates? Researchers need intervals.
Dates can be shifted by a consistent offset per patient, so the time between a patient's studies is preserved while the real dates are not.
What about names burned into the pixels?
Some images, such as ultrasound and scanned documents, carry text drawn into the image itself. We detect and flag or mask burned-in text, so de-identification covers the pixels, not just the header.
Does our data leave our network?
No. De-identification runs on servers inside your environment. Only the de-identified output goes where you decide to send it.
Building a research or AI dataset?
Tell us the source, the size and who will receive the data. We'll design the de-identification with you.