dicomqc

dicomqc audits de-identified DICOM metadata for privacy risks. It reports patient-identifying fields, unexpected pseudonym formats, private tags, and files that could not be read. It never changes the DICOM files.

The desktop app is the recommended interface for selecting data, running audits, and reviewing reports. The CLI supports automation and scripted workflows. Both use the same audit engine.

Two audit modes

Privacy audit inspects one DICOM dataset for identifying metadata. Dataset comparison compares original source DICOM files with their processed candidate copies. It checks file coverage and pseudonym consistency, then applies privacy checks to paired candidate files. It does not compare completed dicomqc runs. Policies and UID checks are optional additions; inventory and MultiQC are optional outputs. See Audit modes for which options each mode supports.

Interface	Privacy audit	Dataset comparison	Options
Desktop App	Worked example	Worked example	Checks and outputs
CLI	Scan walkthrough	Compare walkthrough	Command reference

Project status

dicomqc includes a metadata-only scanner, built-in privacy checks, offline HTML reports, JSON and CSV reports, and MultiQC-compatible scan output. Dataset comparison checks file completeness and patient pseudonym consistency between source and de-identified files. Project policies add site-specific metadata requirements. Scanner inventory optionally lists declared scanner and software labels and private creator blocks. These labels may contain identifying information; review them before sharing. UID integrity checks identifier syntax, role reuse, and conflicting study/series relationships.

Use the desktop app

In Setup, select files or folders. Run an audit, then select Findings or Reports to review its results. Sources and searchable run history remain in the sidebar. File > Save Project stores settings, policies, run history, logs, and generated reports in one .dicomqc file. Original DICOM inputs remain external; working storage is managed by the app.

Load example data, beside DICOM inputs, offers built-in examples without patient data, including an adjustable 1,000- to 100,000-file cohort with deterministic privacy findings generated on demand. Privacy audit is the primary workflow. Policies and UID checks extend it; scanner inventory is optional output. Dataset comparison is a separate mode. The app and its audit service run locally. See Install for macOS, Windows, and Linux packages, then follow the Desktop app guide.

Automate from the command line

Give dicomqc scan one DICOM file or a directory. Directories are scanned recursively:

dicomqc scan candidate_release/

Use --json, --csv, and --multiqc to save reports. Exit codes distinguish a pass (0), warnings that require review (1), and errors or unreadable files (2). See the CLI reference for all options.

Why audit after de-identification?

A de-identification tool can leave identifying metadata behind or miss files. Run dicomqc on its output to check for these problems. You can use the same checks with different tools and data providers.

The recommended process is:

1. Preserve the source DICOM files under restricted access.
2. Pseudonymize or de-identify a working copy with an external tool.
3. Check the resulting files with dicomqc.
4. Review the findings and correct the external tool's configuration.
5. Generate the files again, rerun dicomqc, and retain the final reports.

Terminology

• De-identification is the broader process of reducing the risk that a person can be identified from the data.
• Pseudonymization replaces direct identifiers with a code. The data can still be linked to a person by someone who holds the separate linkage file.
• PHI/PII means identifying or sensitive personal information. Although PHI is a term from US health-privacy law, dicomqc uses it in rule names for identifiable DICOM metadata.
• Value state says whether a tag value is present, empty, or absent. Findings include this state without copying the raw tag value.
• Private tag means a vendor- or organization-defined DICOM data element. Private tags are not automatically safe or unsafe; they require review.

Current checks

The built-in research-release-v0.1 profile checks for:

• metadata fields containing direct identifiers;
• patient identifiers that do not match the expected pseudonym format;
• private DICOM tags that require review;
• unreadable or skipped input files.

YAML policies add requirements for empty fields, required values, approved descriptions, and patient-ID formats. They do not replace the built-in checks.

Reports identify the file, rule, tag, severity, and recommended action for each finding. They also include summary counts. Findings omit observed tag values; see what stays out of reports before sharing.

Limits

dicomqc does not modify files, inspect pixel data or facial features, or certify compliance with DICOM PS3.15, BIDS, HIPAA, or GDPR. A qualified reviewer must still decide whether the data can be shared.

Find what you need

Task	Documentation
Install the CLI	Install
Select files and review audits in the desktop app	Desktop app
Generate the demo and run an audit	Quickstart
Compare source and de-identified datasets	Compare datasets
Add project-specific checks	Project policies
Inspect scanner labels and private creator blocks	Scanner inventory
Check identifiers and study/series relationships	UID integrity
Audit a large MS MRI collection	MS MRI workflow
Review a scan in HTML or MultiQC	Scan walkthrough
Fix reported problems with external tools	Remediation
Understand the implementation	Architecture
Compare related software	Prior work

Documentation map

• Desktop App covers interactive audits, worked examples, and saved projects.
• Install covers setup for both interfaces.
• CLI covers commands, optional checks, and scripted workflows.
• Data preparation covers external remediation and the MS MRI workflow.
• Technical Details explains the implementation.
• About records citation guidance, prior work, and the project disclaimer.

Project development and issue tracking take place in the dicomqc GitHub repository.