Licensing and attribution

All data in IDC is publicly available: no registration, no access requests. "Publicly available" is not the same as "unrestricted", though. Every DICOM series in IDC carries a license - most often a Creative Commons license - that defines what you may do with that series and how you must credit the people who produced it.

This page covers which licenses you will encounter in IDC, why the license is a property of the individual series rather than of the collection, how to check the license for any selection of data, and how to attribute what you used.

Counts and version numbers on this page are as of IDC data release v24, and will change as new data is added. Nothing on this page is legal advice - when in doubt, read the license text linked below and consult your institution.

Licenses used in IDC

License	Commercial use	Series	Collections	Share of IDC by size
CC BY 4.0	allowed	865,935	126	74.7%
CC BY 3.0	allowed	132,303	79	22.1%
CC BY-NC 4.0	not allowed	28,783	5	2.1%
CC BY-NC 3.0	not allowed	5,851	4	0.8%
NLM Terms and Conditions; May 21, 2019	see terms	39	1	0.3%

Altogether that is 1,032,911 series in 176 collections and 94.7 TB. About 97% of the data by size is under a plain CC BY license that permits commercial reuse; just under 3% is non-commercial. The collections column sums to more than 176 because a single collection can contain series under several licenses - see the next section.

IDC hosts only data that its source has released under a license permitting public redistribution. Where a source collection is split across licenses and part of it is restricted, that part is not in IDC at all: the radiology component of TCGA-GBM is covered by the TCIA limited access license and is absent from IDC, while its digital pathology component is CC BY and is present (see the v9 entry in Data release notes).

Every license used in IDC - CC BY and CC BY-NC alike - requires attribution. See Attribution and citation below for how to produce it.

The license belongs to the series, not to the collection

license_short_name is a series-level attribute, exactly like source_DOI (see Data model). Do not assume that the license of a collection tells you the license of everything you selected from it: 39 of the 176 collections in IDC carry more than one license.

That is because a collection is not licensed as a whole. It is composed of one or more contributing sources, each bringing its own DOI, license and citation - collections_index records them in its sources field, tagged original_data or analysis_result. A collection carries several licenses whenever its sources do, which happens in two ways.

A contributed analysis result brings its own license. An analysis result is a collection in its own right, but its series keep the collection_id of the images they analyze - so they sit inside the original collection under different terms. In NSCLC-Radiomics, the original CT images are CC BY-NC 3.0, while the AI-derived annotations added later are CC BY 4.0:

license_short_name  series  size_GB
         CC BY 4.0    3661      9.3     <- nnu_net_bpr_annotations (analysis result)
      CC BY-NC 3.0    1265     34.9     <- original CT images

A collection can have several original sources too. NLST is assembled from two original_data sources released under different terms: the CT images from TCIA (10.7937/tcia.hmq8-j677, 203,087 series, CC BY 4.0) and the DICOM-converted slide microscopy images (10.5281/zenodo.12689650, 1,259 series, CC BY 3.0). No analysis result is involved here - this is simply what the collection is made of.

The bucket name is not a license label. It is tempting to read idc-open-data-cr as "the non-commercial bucket", but the mapping is not reliable: all 4,433 CC BY-NC 3.0 series of the Phantom FDA collection are stored in idc-open-data. Always read license_short_name from the metadata - never infer the license from aws_bucket or from the URL you downloaded from.

A permissive license on a derived object does not relicense the images it describes. The bamf_aimi_annotations analysis result contributes 897 CC BY 4.0 segmentation series to Duke-Breast-Cancer-MRI, whose 5,161 MR series are CC BY-NC 4.0. The segmentations are of little use without the images, and using the two together means satisfying both licenses - the more restrictive one governs the combination.

The license is not inside the DICOM file. It is metadata that IDC maintains alongside the files, not a DICOM attribute, so it does not travel with the .dcm files you download. Nothing is lost, though: the SeriesInstanceUID is in the file, and it is the only thing you need to retrieve the license again at any later time - through idc-index, an AI assistant, the REST API or BigQuery, as shown below. Prefer looking it up to relying on a value you recorded earlier: license assignments are among the mutable metadata IDC may revise between releases - see mutable_metadata in BigQuery tables.

Checking the license for your selection

Every IDC interface can answer this. They are listed here from fewest prerequisites to most.

IDC Portal

Nothing to install: the Explore page lets you filter data by license type, so you can restrict a cohort to commercially reusable data before building a manifest.

idc-index

The index table has one row per series, with license_short_name among its columns:

from idc_index import IDCClient

client = IDCClient()
client.sql_query("""
SELECT license_short_name, COUNT(*) AS series, ROUND(SUM(series_size_MB)/1024, 1) AS size_GB
FROM index
WHERE collection_id = 'nsclc_radiomics'
GROUP BY license_short_name
ORDER BY series DESC
""")

The same table resolves a file you already have on disk back to its terms, given only the SeriesInstanceUID you can read out of it:

SELECT license_short_name, source_DOI
FROM index
WHERE SeriesInstanceUID = '1.3.6.1.4.1.14519.5.2.1.7695.4164.174071765480311650274095134055'

To keep only data you can use commercially, filter on an explicit allowlist rather than excluding the restrictions you happen to know about:

SELECT SeriesInstanceUID, collection_id, license_short_name, series_aws_url
FROM index
WHERE Modality = 'CT'
  AND license_short_name IN ('CC BY 3.0', 'CC BY 4.0')

AI assistants

Ask the assistant for the license breakdown of your cohort before you download it; the get_licenses tool answers exactly the question below. See Using IDC with an AI assistant.

REST API

POST /v3/licenses returns the license breakdown for any cohort filter:

curl -s https://api.imaging.datacommons.cancer.gov/v3/licenses \
  -H 'Content-Type: application/json' \
  -d '{"terms": {"collection_id": ["nsclc_radiomics"]}}'

{"licenses":[{"license_short_name":"CC BY 4.0","series":3661,"size_TB":0.01},
             {"license_short_name":"CC BY-NC 3.0","series":1265,"size_TB":0.036}]}

The response summarizes the series matched by the filter, so you can check the license at any granularity - a whole collection, or a single study or series by its UID. See Getting the data.

BigQuery

Listed last because it is the only option here that needs a Google Cloud project of your own to run queries under - but it is also the most detailed. BigQuery gives you the license at the granularity of the individual instance, with the full name and the URL of the license text: license_short_name, license_long_name and license_url are columns of the dicom_all and auxiliary_metadata tables. See BigQuery tables.

Collections that are not entirely CC BY

As of v24, ten collections contain data under a license other than plain CC BY:

Collection	collection_id	License	Series	Patients
Breast-Cancer-Screening-DBT	breast_cancer_screening_dbt	CC BY-NC 4.0	22,032	5,060
Duke-Breast-Cancer-MRI	duke_breast_cancer_mri	CC BY-NC 4.0	5,161	922
MIDRC-RICORD-1A	midrc_ricord_1a	CC BY-NC 4.0	229	110
MIDRC-RICORD-1B	midrc_ricord_1b	CC BY-NC 4.0	120	117
MIDRC-RICORD-1C	midrc_ricord_1c	CC BY-NC 4.0	1,241	361
NSCLC-Radiomics	nsclc_radiomics	CC BY-NC 3.0	1,265	422
NSCLC-Radiomics-Genomics	nsclc_radiomics_genomics	CC BY-NC 3.0	89	89
NSCLC-Radiomics-Interobserver1	nsclc_radiomics_interobserver1	CC BY-NC 3.0	64	22
Phantom FDA	phantom_fda	CC BY-NC 3.0	4,433	7
NLM-Visible-Human-Project	nlm_visible_human_project	NLM Terms and Conditions	39	2

The counts above are for the restricted series only. Duke-Breast-Cancer-MRI and NSCLC-Radiomics additionally contain CC BY 4.0 analysis results, which is why they also appear among the mixed-license collections.

Treat this table as a snapshot, not as a check you can rely on. Licenses are assigned per series and can be added or revised in any release - query license_short_name for your own selection rather than filtering on this list of collection names.

Attribution and citation

Attribution is a condition of every license in IDC, and - like the license itself - it attaches to the series. Each series records the source it came from in source_DOI, and that source, not IDC and not the collection, is what you cite. So the citations you owe are determined by the set of source_DOI values present in your selection, whatever shape that selection has:

SELECT source_DOI, license_short_name, COUNT(*) AS series
FROM index
WHERE <your selection>
GROUP BY source_DOI, license_short_name

The same selection can span several DOIs and several licenses at once. Even selecting a single patient can do it: in the PROSTATEx example in Data model, one patient's study spans five DOIs from four contributing groups, under two different licenses.

You do not have to assemble the citations by hand. POST /v3/citations (or the get_citations tool, if you are working through an AI assistant) resolves the source_DOI values of the series matched by your filter into formatted citations. The filter below happens to name a collection, but any cohort filter works - down to a single SeriesInstanceUID:

curl -s https://api.imaging.datacommons.cancer.gov/v3/citations \
  -H 'Content-Type: application/json' \
  -d '{"filters": {"terms": {"collection_id": ["nsclc_radiomics"]}}, "citation_format": "apa"}'

This selection covers two DOIs, so two citations come back - one for the original images, one for the AI-derived annotations that were added to the same collection:

Aerts, H. J. W. L., Wee, L., Rios Velazquez, E., et al. (2019). Data From NSCLC-Radiomics (Version 4) [Dataset]. The Cancer Imaging Archive. https://doi.org/10.7937/K9/TCIA.2015.PF0M9REI

Deepa Krishnaswamy, Dennis Bontempi, David Clunie, Hugo Aerts, & Andrey Fedorov. (2023). AI-derived annotations for the NLST and NSCLC-Radiomics computed tomography imaging collections [Dataset]. Zenodo. https://doi.org/10.5281/ZENODO.7473970

Had you selected only the CT images, only the first citation would apply - and the non-commercial restriction along with it. apa, bibtex, csl-json and turtle formats are supported.

In addition to the per-source citations, please acknowledge IDC itself by citing the IDC overview publication:

Fedorov, A., Longabaugh, W. J. R., Pot, D., et al. National Cancer Institute Imaging Data Commons: Toward Transparency, Reproducibility, and Scalability in Imaging Artificial Intelligence. RadioGraphics (2023). https://doi.org/10.1148/rg.230180

Part 3 of the "Getting started" tutorial walks through licenses, DOIs and attribution hands-on.

Related pages

• Data model - why licensing and provenance attach at the series level

• Files and metadata - buckets and metadata sources

• Data versioning - how data, and its mutable metadata, change between releases

• Getting the data - licenses and citations via the REST API