dicom-metadata-extract

DICOM Metadata Extract is a small, self-contained NVIDIA MedTech skill that extracts selected metadata from a single DICOM medical-imaging file and flags the presence of standard-tag PHI (protected health information). It is deliberately narrow: it reads one file with pydicom and emits JSON to stdout containing fields like transfer syntax, modality, grouped study/series/image metadata, a phi_present flag, and the list of PHI tags found. It is explicitly not for anonymization, de-identification, or clinical use, and serves as the reference "smallest end-to-end" example skill for the Medical AI Skills layout.

The workflow is a documented wrapper: the agent reads skill_manifest.yaml before changing arguments, runs scripts/extract_metadata.py (either directly via Bash/Python or through a host `run_script` interface), keeps outputs under a caller-provided run directory, and can validate results against an output schema and an optional quality verifier for evidence review. It extracts only literal DICOM header values and does not infer modality, body part, diagnosis, or clinical meaning. Documented limitations are clear: it covers only a small PS3.15-inspired standard-tag subset, does not check private tags, does not detect burnt-in pixel PHI, has minimal multi-frame handling, and is not for clinical deployment, regulatory de-identification, autonomous diagnosis, or patient-facing use.

Target users are developers and engineers doing development-time review of DICOM files — inspecting header metadata and getting a coarse PHI-presence signal. It is research-and-development only, requires the pydicom dependency, and is designed to be run from the repository root.