TCIA Query Skill

tcia-query-skill helps AI agents find, verify, cite, and access datasets published by The Cancer Imaging Archive (TCIA).

It uses TCIA’s WordPress Collection Manager as the publication authority, TCIA’s Publications EndNote XML as the verified record of papers that analyzed TCIA data, and a release-backed SQLite snapshot as its normal discovery layer. It then routes users to the appropriate data system, such as IDC, CTDC, General Commons, PathDB, DataCite, TCIA Data Retriever, or Aspera.

The skill brings metadata and provenance from authoritative source systems into a verified release bundle, then supports different use cases through skill guidance, MCP, REST, or local artifacts. The underlying imaging and supporting data remain with the systems responsible for them.

Participant aggregates, harmonized clinical values, inferred labels, identity groupings, and computed counts represent an additional layer of data curation derived from the original datasets. The bundle and its MCP/REST projections are discovery and analysis aids, not authoritative replacements for those datasets. Verify authoritative dataset details on the TCIA WordPress Collection or Analysis Result page and in its linked original supporting files. Validation of a bundle does not establish source completeness or clinical correctness; discrepancies should be checked against the sources and reported explicitly.

What It Can Do

• Find TCIA Collections and Analysis Results by disease, body site, modality, data type, program, access level, license, DOI, or supporting data.
• Distinguish newly published datasets from datasets updated recently.
• Find current downloads, clinical metadata, annotations, segmentations, pathology slides, NIfTI files, and related Analysis Results.
• Search TCIA’s verified analytical-use library for papers that analyzed TCIA datasets.
• Retrieve DataCite DOI metadata, author lists, event metrics, and formatted dataset citations.
• Route public DICOM through IDC and create viewer links or portable TCIA Data Retriever manifests.
• Explain controlled-access requirements without treating a public metadata record or manifest as authorization.

Example prompts:

• “Find TCIA datasets with breast MRI and tell me how to access them.”
• “Which datasets were newly published or updated recently?”
• “Does this Collection have segmentations in a related Analysis Result?”
• “Find peer-reviewed papers using this TCIA dataset DOI.”
• “Create a TCIA Data Retriever CSV for these Series Instance UIDs.”
• “Is this dataset open, noncommercial, mixed, or controlled access?”

Try The Hosted Demo

A read-only demonstration server exposes the same snapshot-backed query layer: