Six questions before buying
A promising preview cannot answer whether a dataset supports the roadmap. Enterprise diligence establishes what patient or examination unit is counted, which sequences or reconstructions are complete, how labels were produced, which populations and scanners are represented, what was excluded and what the licence permits.
- Does the imaging match the model input and intended claim?
- Are positives, controls and difficult cases defined consistently?
- Is ground truth suitable for the task and independently reviewable?
- Are site, scanner and demographic distributions documented?
- Has de-identification been validated across metadata and pixels?
- Can the supplier prove authority for the offered commercial use?
Start with a pilot, not a full transfer
A small sample should exercise the real acceptance criteria: file structure, sequence completeness, label mapping, image quality, metadata preservation and privacy controls. Pilot failures are cheaper to correct before the final cohort is curated.
Price follows work and rights
Case count alone does not determine value. Rare pathology, longitudinal linkage, expert segmentation, multi-site coverage, strict exclusivity and complex privacy work can dominate cost. A defensible quote separates data access, curation, clinical review, privacy verification and licensing scope.
Questions, answered directly.
What is the difference between a scan and a DICOM file?+
A scan or examination can contain many series and hundreds or thousands of DICOM files. Programme scale should be expressed in verified studies, not raw file counts.
What makes a dataset valuable?+
Task fit, rare or representative cohorts, reliable ground truth, technical completeness, privacy evidence and clear commercial rights.
When is exclusivity appropriate?+
Only when exclusivity creates material strategic value. It narrows future licensing and usually increases cost.