Africa's medical imaging archives.Made AI-ready.
Blimdata turns archived hospital scans into de-identified, radiologist-annotated training datasets — so global AI teams can prove their models work on Black patients, and African hospitals finally earn from data they already own.
Dataset output
- PHI stripped
- 18 / 18
- Annotation
- Radiologist
- Format
- DICOM + JSON
- Provenance
- Attested
The market gap
Two problems. One missing bridge.
Global AI is blocked at the approval gate.
Regulators increasingly expect evidence that a diagnostic model performs across the populations it will be used on. Without imaging data from Black patients, that evidence cannot be produced — and global expansion stops.
- Diagnostic algorithms trained on narrow populations degrade on the ones they omit.
- Demographic coverage has become a submission requirement, not a nice-to-have.
- Buying diverse imaging data at scale is, today, close to impossible.
African archives sit idle on local servers.
Millions of historical X-rays, CTs and MRIs are already stored on hospital PACS across the continent. They generate no revenue, because the institutions holding them have no way to anonymise, annotate and package them safely.
- No engineering capacity to extract and process DICOM at scale.
- No compliance tooling to prove de-identification to an international buyer.
- No commercial route to the companies that need the data most.
The pipeline
From hospital PACS to training-ready dataset.
We do not broker folders of raw files. Every study passes through four controlled stages before a licensee ever sees it — which is what turns a low-margin data handoff into a premium AI asset.
Secure ingestion
A lightweight gateway runs inside the hospital network and pulls studies from the existing PACS over standard DICOM query/retrieve. Read-only, outbound-only, no new inbound firewall rules.
Automated de-identification
All 18 PHI identifiers are stripped from DICOM headers, and burned-in text is detected and redacted from pixel data. Studies that fail validation never leave the building.
Expert annotation
Licensed radiologists label pathology in a secure workspace. AI pre-segmentation proposes masks and bounding boxes; the clinician confirms, corrects or rejects each one.
Governed licensing
Versioned datasets are held in a controlled data room and licensed by subscription or batch, with provenance records and de-identification attestations attached to every release.
Architecture
PHI never leaves the hospital.
The gateway that reads your PACS is the same process that strips the identifiers. Re-identifiable data is never transmitted, never staged in our cloud, and never held by us — because the de-identification step runs before anything touches the wire.
This is not a policy setting that somebody could switch off. It is where the software runs.
Who we serve
Two sides. One pipeline.
Your archive is already an asset. We make it earn.
Turn scans you are already storing and already paying to keep into a recurring revenue line — without adding headcount, changing clinical workflow, or letting identifiable data off your premises.
- Zero cost to deploy. We fund the gateway, the processing and the annotation.
- Revenue share on every licence attributable to your data.
- Read-only PACS integration; radiology workflow is untouched.
- Written data agreement before a single study is processed.
Diverse imaging data, with the paperwork regulators ask for.
Acquire annotated, provenance-tracked studies from populations your current training set does not represent — packaged as structured datasets, not as a folder of files you have to clean yourself.
- DICOM plus structured annotation and clinical metadata.
- Provenance and de-identification attestation on every release.
- Subscription, batch acquisition or commissioned collection.
- Sample datasets released under NDA before you commit.
Dataset coverage
Modalities in the pipeline.
Coverage grows with every partner hospital we onboard. Current and committed volumes are shared under NDA rather than published.
X-ray
Chest, musculoskeletal and abdominal plain film — the highest-volume archive in most partner sites.
Computed tomography
Head, chest and abdominal series, including multi-phase contrast studies.
Magnetic resonance
Neuro and musculoskeletal sequences, T1 / T2 / FLAIR and diffusion.
Ultrasound
Obstetric, abdominal and cardiac studies, including cine loops.
Mammography
Screening and diagnostic views, with priors where available.
Commissioned collection
Where an archive does not yet hold what you need, we can scope prospective collection with partner sites.
Volumes and pathology breakdowns are shared under NDA.
Common questions
Clear answers.
The questions hospital boards and data-acquisition teams ask us first.
Do patients consent to this?
Partner hospitals operate under their own institutional ethics and consent frameworks, and remain the data controller throughout. We only process studies an institution is lawfully permitted to release, every study is de-identified inside their network before transfer, and nothing is processed before a written agreement is in place. We do not accept identifiable data.
What exactly gets removed from a scan?
All 18 protected health information identifiers defined by the HIPAA Safe Harbor method are stripped from DICOM metadata — names, medical record and accession numbers, device serial numbers, exact dates, geography below state level, contact details and free-text fields. Separately, burned-in text rendered into the pixel data itself is detected and redacted, because header cleaning alone does not catch it.
Does the gateway open a hole in our firewall?
No. The gateway establishes outbound TLS 1.3 connections only. There are no inbound ports, no VPN tunnel and no new firewall exceptions. It reads from PACS using standard DICOM query/retrieve under a read-only application entity title, so it cannot write to or modify your archive.
What does a hospital pay?
Nothing. Blimdata funds deployment, processing, annotation and compliance work. Partner institutions receive a share of the licensing revenue attributable to their data, reported against an auditable record of what was licensed.
How is data licensed to buyers?
Either as a recurring subscription to a versioned dataset that grows over time, or as a one-off batch acquisition. Terms cover permitted use, redistribution, model-training rights, derived-work ownership and retention. Exclusive and semi-exclusive arrangements are available for defined modality and pathology combinations.
Where is the data stored, and who can reach it?
De-identified studies and annotations are held in encrypted cloud storage — AES-256 at rest, TLS 1.3 in transit — with configurable regional data residency. Access is role-based and least-privilege, and every read, export and permission change is written to an immutable audit log.
Are you compliant with the Nigeria Data Protection Act?
The pipeline is built to the NDPA, and de-identification follows the HIPAA Safe Harbor standard so that resulting datasets are also usable by licensees operating under US and EU regimes. Compliance documentation is provided to partner institutions and to licensees as part of diligence.
How long does it take to onboard a hospital?
A typical engagement starts with a written assessment of the archive, then a technical review with your PACS administrator, then a data agreement. Gateway deployment itself is measured in days rather than months, because it requires no changes to your existing systems.
Start with a single archive.
One department, one modality. We will show you exactly what comes out of the other end — and what it is worth — before you commit to anything.