Independent Validation for Medical Imaging AI Datasets
Know whether your dataset is audit-defensible — before it costs you.
Princeton Medical Systems validates medical imaging datasets against VIDS — the open Verified Imaging Dataset Standard. We issue a signed Validation Attestation that holds up in FDA submissions, procurement audits, and vendor disputes. Independent. Reproducible. Final.
Request an EvaluationThe Service
What a Compliance Evaluation Is
A Compliance Evaluation is an independent assessment of a medical imaging dataset against VIDS — the open standard our team authored and published under CC BY 4.0. Princeton Medical Systems runs the validator, reviews edge cases that the validator flags conservatively, and issues a Validation Attestation: a signed third-party document confirming whether the dataset meets the standard.
Acceptance decisions — whether to acquire a dataset, whether to include it in an FDA submission, whether to accept a vendor delivery — depend on criteria. PMS defines those criteria, through VIDS. The Compliance Evaluation is how those criteria get applied to your dataset.
What You Receive
The Deliverables
PASS / FAIL Determination
Across all 21 VIDS validation rules. Binary, reproducible, no ambiguity.
Validation Report
Machine-readable JSON capturing every rule, every result, every flagged record. The underlying evidence buyers and auditors can re-verify independently.
Validation Attestation
Signed PDF issued by Princeton Medical Systems with a named radiologist or engineer attached. Dated, version-stamped, designed to hold up in regulatory and procurement settings.
Edge-Case Review Notes
For any rule the validator flags conservatively — explicit defect-vs-acceptable-limitation judgment from a credentialed reviewer. This is what a self-run validator output cannot produce.
How It Works
The Process
Submit dataset details and scope via the form below or by emailing info@princetonmedicalsystems.com.
We confirm scope, send a quote (typically within 1 business day), and define the turnaround.
Secure data transfer with a 50% deposit. We accept de-identified DICOM, NIfTI, and standard structured exports.
The VIDS validator runs against the full dataset. 21 rules, executed automatically. Hours, not days.
A credentialed radiologist or engineer reviews any rule failures, distinguishing real defects from acceptable historical limitations. Each judgment is documented in the Report.
Validation Attestation (signed PDF) plus Validation Report (JSON) delivered to you. Final 50% payment on delivery.
Use Cases
When a Compliance Evaluation Is Worth It
You're acquiring imaging datasets from external vendors and need third-party verification of what you're paying for — before money changes hands.
You're preparing an FDA or CE submission and need audit-defensible documentation of your training data's structure, provenance, and quality.
You're a vendor delivering imaging datasets and your buyer's contract requires independent validation, not a self-run validator output.
You can't afford the cost or timeline of dataset rework after a regulator or auditor finds problems your team missed.
Independence
Why You Can Trust the Result
The standard is open. VIDS is published under CC BY 4.0 at vidsstandard.org. The reference validator is open-source on PyPI. You can read the spec, install the validator, and run it yourself — before, during, or after our evaluation. Our methodology is not a black box.
The validation is reproducible. Every Validation Report is version-stamped against a specific validator release. The exact same validator, run against the exact same dataset, produces the exact same result — in your environment, in our environment, in an auditor's environment.
The attestation is independent. Princeton Medical Systems has no commercial interest in the third-party datasets we evaluate — we are not the seller, broker, or producer of those datasets. For evaluations on PMS's own reference datasets, the same reproducible validator output applies, and buyers can independently verify.
The Stakes
Why This Matters
When VIDS is run against widely-used public datasets, results are sobering: BraTS scores 39%, MSD 30%, LIDC-IDRI 27%, CheXpert 20%. These are reference datasets. If they score this low, the question for any buyer is what's in the private vendor delivery they're about to accept.
Defects discovered after acquisition cost more to fix than defects caught before purchase. The cost is paid in rework, in regulatory delays, and in the procurement disputes that follow.
In most cases, the cost of a Compliance Evaluation is lower than the cost of correcting a dataset after acquisition.
Engagement
How Engagements Are Scoped
Compliance Evaluations are scoped individually. Dataset complexity varies materially — modality, study count, annotation density, provenance documentation, and turnaround requirements all affect scope. We quote each engagement against the actual dataset rather than against generic tiers.
Submit your evaluation request through the form below or by emailing info@princetonmedicalsystems.com. We respond within one business day with a quote that includes scope, deliverables, turnaround, and pricing. Most Compliance Evaluations complete within 5 business days from secure data transfer.
Engagement structure: 50% deposit on scope agreement, 50% on delivery of the Validation Attestation and Validation Report.
Questions
Common Questions
How is this different from running the open-source validator myself?
The validator output is binary PASS/FAIL on 21 rules — useful, but not sufficient on its own. A Compliance Evaluation adds independent attestation that holds up in regulatory submissions, expert review of edge cases the validator flags conservatively, dispute-resolution authority when buyers and sellers disagree, and the evidentiary workflow that turns a JSON file into something usable in a procurement decision. The free validator drives adoption. The paid Compliance Evaluation converts adoption into something defensible.
Can I see a sample Validation Report and Attestation?
Yes. Sample artifacts available on request. The LIDC-Hybrid-100 reference dataset on Zenodo includes a public Validation Report demonstrating what the JSON output looks like.
What happens if my dataset fails?
The Validation Report identifies every failed rule with documented reasoning. You can remediate the underlying issues and request re-evaluation, restructure the procurement on the basis of the failure, or proceed with the dataset documented as non-compliant — your call. PMS is not a gatekeeper; we issue the documented evidence on which acceptance decisions are made.
Do you accept datasets in formats other than DICOM?
Yes. NIfTI, structured exports with documented provenance, and standard medical imaging formats. Submit the format details with your evaluation request and we'll confirm scope.
How is data security handled?
De-identified datasets only. Secure transfer protocols, encrypted storage during evaluation, and configurable retention policies. Specific security controls quoted as part of the engagement scope.
Is the underlying validator FDA-cleared?
VIDS is an open standard for dataset validation, not a regulated medical device. The validator does not make clinical or diagnostic claims; it assesses dataset structure, provenance, and quality documentation. Validation Attestations are designed to be referenced in regulatory submissions as evidence of training data due diligence — not as device clearance themselves.
Ready to Evaluate Your Dataset?
Tell us about the dataset, the use case, and the timeline. We'll respond with a quote and scoped engagement plan within one business day.
Request an EvaluationOr email info@princetonmedicalsystems.com directly.