PMS Imaging Standard™
The Gold Standard for Medical Imaging Datasets
Every dataset is validated against VIDS — the open Verified Imaging Dataset Standard authored by our team. Clinically validated, fully traceable, bias-aware, and compliance-ready.
The Challenge
Medical Imaging AI Has a Data Problem
Most medical imaging datasets used for AI training lack standardized validation, traceability, and regulatory alignment.
This creates compounding risk: models trained on unverified data face regulatory rejection, bias issues, and clinical safety concerns. The cost of poor data isn't just technical — it's measured in delayed treatments, failed submissions, and eroded trust.
Framework
Four Pillars of Validation
Every dataset validated against the VIDS standard meets rigorous requirements across four critical dimensions.
Clinical Validation
Every annotation reviewed and signed off by credentialed radiologists. Annotator credentials are documented in the dataset's provenance record. No automated-only labels.
Full Traceability
Complete data provenance from source through every transformation. Every change logged, every origin verified.
Bias Awareness
Systematic demographic and acquisition protocol diversity checks to identify and document representation gaps.
Regulatory Compliance
Aligned with FDA and CE Mark requirements for AI/ML medical devices. Audit-ready documentation included.
Process
How It Works
From raw DICOM data to validated, ML-ready export — a rigorous five-stage pipeline.
Validation Engine
What the Validator Checks
Our validation engine runs comprehensive checks across six critical dimensions to ensure dataset integrity.
Metadata Completeness
All required DICOM tags present and correctly formatted
Annotation Consistency
Label formats, coordinate systems, and terminology aligned
Demographic Representation
Statistical checks for age, sex, and ethnicity distribution
DICOM Tag Integrity
Tag values validated against VR definitions and standards
De-identification Verification
PHI removal confirmed across all headers and pixel data
Provenance Chain Completeness
Full audit trail from original source through all transformations
Compatibility
Integration & Export
Datasets validated against VIDS export directly into your preferred ML framework.
NVIDIA's open-source framework for healthcare AI
Self-configuring segmentation framework
Datasets library for ML research
Comparison
VIDS-Validated vs. Typical Datasets
| Dimension | VIDS-Validated | Typical Dataset |
|---|---|---|
| Clinical Validation | ✓ Radiologist-verified | ✗ Automated or unverified |
| Traceability | ✓ Full provenance chain | ✗ Partial or missing |
| Bias Assessment | ✓ Systematic diversity checks | ✗ Rarely documented |
| Regulatory Alignment | ✓ FDA/CE ready | ✗ Not addressed |
| Export Compatibility | ✓ MONAI, nnU-Net, HF | ~ Variable formats |
| Annotation Consistency | ✓ Standardized schema | ✗ Inconsistent labeling |
Research & Open Source
Built on an Open, Peer-Reviewed Standard
The PMS Imaging Standard is our commercial implementation of VIDS — the Verified Imaging Dataset Standard, an open specification authored by our team. The specification is published under CC BY 4.0, the reference validator is Apache 2.0 on PyPI, and our reference dataset is freely available on Zenodo. This means buyers can independently verify the methodology behind every dataset we deliver.
Muthu & Shalen. VIDS: A Verified Imaging Dataset Standard for Medical AI. 21 machine-enforceable validation rules across two compliance profiles.
Apache 2.0. CI-ready with JSON output and standard exit codes. Browser-based drag-and-drop validation also available at vidsstandard.org.
Published on Zenodo with a permanent DOI. Demonstrates a fully VIDS-compliant lung CT dataset with structured provenance and quality documentation.
Start Building with Validated Data
Request a sample dataset validated against the VIDS standard
Request Sample Dataset