Your healthcare AI training data partner.

De-identified clinical text and EHR records, medical imaging across CT, MRI, X-ray, ultrasound and pathology, clinical audio and physician dictation, physiological signals such as ECG and EEG, claims and administrative data, and linked multimodal datasets. Availability depends on specialty, geography and intended use, so we run a feasibility assessment against your specification before committing to volumes.

We work to a named standard rather than best effort. The default is the HIPAA Safe Harbor method, removing all 18 specified identifiers across free text, DICOM headers, burned-in pixel data and audio. Where Safe Harbor removes information your model genuinely needs — precise dates in a longitudinal study, for example — we use Expert Determination, where a qualified statistician certifies that re-identification risk is very small. Either way you receive documentation of the method applied and the residual risk assessment.

Properly de-identified data falls outside HIPAA’s definition of protected health information, which is what makes it usable for model training. Where an engagement requires us to handle identified PHI — during a de-identification project, for instance — we operate under a Business Associate Agreement with the covered entity. We will walk your compliance team through the specific arrangement for your engagement.

Yes, where the underlying source agreement permits commercial AI training, and we will tell you plainly when it does not. Licensing terms cover permitted use, derivative model rights, term and territory, and we provide provenance documentation for each source so your legal team can assess the chain of rights rather than take a warranty on faith.

Trained annotators working under clinical supervision, with practising radiologists, physicians and certified medical coders doing review and adjudication on specialty work. Every annotator signs an NDA and completes confidentiality and protocol training. We report inter-annotator agreement per batch so you can see label quality as a number rather than a promise.

Most clinical AI falls under the EU AI Act as high-risk, which brings Article 10 data-governance duties: training, validation and test sets must be relevant, sufficiently representative, as free from errors as possible, and examined for bias. We prepare datasets against those criteria and supply the documentation — demographic composition, known gaps, bias assessment — that forms part of your technical file. Obligations begin applying from December 2027 for Annex III systems.

It depends on modality, volume, annotation complexity and the compliance regime in scope — a single-label image classification set and an adjudicated multi-organ segmentation set are different projects at different prices. After a scoping call we provide a written specification with fixed pricing per deliverable, so you are not signing an open-ended time-and-materials arrangement.

Pilot batches typically move in weeks rather than months, with the variable being compliance review and source availability rather than production capacity. We would rather set a realistic date after the scoping call than quote a number here that ignores your specialty and jurisdiction.