Medical imaging data for AI training
Radiology (X-ray, CT, MRI, ultrasound), pathology slides, dermatology, ophthalmology and dental images with reports or labels.
- Measured in
- Studies, images or slides, by modality, body part and finding.
- Who buys
- Medical AI companies, device makers and research groups; increasingly general labs adding health capability.
- Where it comes from
- Hospitals and health systems, imaging and diagnostic centers, specialist practices, dental groups and research archives.
- Rights to clear
- HIPAA, state health privacy laws, patient consent forms, research protocols and data use agreements.
- Privacy
- Patient identifiers in DICOM metadata, burned-in annotations and facial features in head scans; HIPAA de-identification required.
- Pricing signal
- Priced by study, modality and label quality. Expert-labeled rare findings command the most.
- Medical imaging datasets train diagnostic and triage models, and labels from qualified clinicians are what make them valuable.
- Under HIPAA, imaging data must be de-identified by safe harbor or expert determination before it can be licensed outside a business associate relationship.
- Identifiers hide in DICOM headers and burned into pixels, so de-identification covers both.
- Health systems, imaging centers, dental and veterinary practices and research groups are the usual holders.
HIPAA first
Protected health information can only be shared outside HIPAA's permitted uses if it has been de-identified. The US Department of Health and Human Services describes two methods: safe harbor, which removes 18 types of identifiers, and expert determination, where a qualified expert certifies the risk of re-identification is very small (HHS). Imaging deals usually use expert determination because it can preserve useful dates and details that safe harbor would strip.
Where identifiers hide in images
- DICOM header fields (names, IDs, dates, institution, referring physician)
- Text burned into pixels by the scanner or annotations
- Facial features reconstructable from head CT and MRI
- File names and folder structures
What makes imaging data valuable
- Labels or reports from qualified clinicians, ideally with structured findings
- Rare conditions and underrepresented populations
- Follow-up studies showing outcomes
- Multiple scanner vendors and sites, which helps models generalize
Deal structure
Health data deals often involve a data use agreement, a de-identification certificate and limits on re-identification and linking. Some providers license only through a research partnership. We help you work out which route fits before any data is prepared.
Questions
Can a practice sell patient images?
Not as identifiable patient data. Images must be de-identified under HIPAA before licensing, and state law and patient consent forms also apply. Many deals use an expert determination.
Are radiology reports included?
Often, after de-identification. Reports supply the labels that make images useful for training.
Can veterinary images be licensed?
Yes, and HIPAA does not apply to animal records, though owner details still need removing.
Related guides
Related data types
Sources
Need medical imaging data, or hold it?
Tell us what you need or what you hold. We check rights and privacy first, then introduce vetted counterparties. You license directly with each other.
- We reply to every enquiry within two working days.
- No upfront fees for sellers
- We never hold your raw data