Written and maintained by CASRAI Editorial Board
Last updated
Quick answer: The Medical Imaging and Data Resource Center (MIDRC) is an open-access medical imaging data commons founded in 2020 to foster machine intelligence in healthcare. Funded by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under contract 75N92020D00021 and supported by ARPA-H, MIDRC is co-led by the Radiological Society of North America (RSNA), the American College of Radiology (ACR), and the American Association of Physicists in Medicine (AAPM), with data hosted on the Gen3 platform at the University of Chicago.
What is MIDRC?
The Medical Imaging and Data Resource Center (MIDRC) is a multi-institutional research initiative that curates, harmonizes, and distributes clinical imaging datasets for machine learning applications. Operating within biomedical engineering research, MIDRC bridges clinical radiology and algorithm development by providing access to verified, de-identified clinical data adhering to FAIR data principles (Findable, Accessible, Interoperable, and Re-usable).
MIDRC was created to resolve a central challenge in medical imaging AI: the lack of diverse, multi-institutional datasets with standardized clinical metadata. Local machine learning models trained on single-hospital data often degrade when deployed across different patient populations or scanner manufacturers. By aggregating diagnostic studies from academic medical centers and community hospital systems, MIDRC supports robust, generalizable algorithmic development.
In addition to its open repository, MIDRC partners with federal computing initiatives including the ARPA-H Biomedical Data Fabric (BDF) and the National Artificial Intelligence Research Resource (NAIRR) pilot. These pipelines assist researchers operating under design controls medical device industry workflows and regulatory standards for software evaluated as a medical device under what is a medical device definitions.
Who runs it and what it covers
MIDRC is administered through a collaborative governance model uniting three primary professional societies:
- Radiological Society of North America (RSNA): Leads clinical curation, community engagement, and data intake from radiology departments.
- American College of Radiology (ACR): Oversees registry integration, clinical data governance, and data contribution pipelines.
- American Association of Physicists in Medicine (AAPM): Governs image quality assurance, calibration metrics, harmonization protocols, and benchmarking standards.
- University of Chicago: Hosts and operates the primary data portal utilizing the open-source Gen3 cloud platform.
- Funding oversight: Supported by the NIH NIBIB under contract 75N92020D00021, with additional support from ARPA-H.
Initially launched to address COVID-19 thoracic imaging, MIDRC has expanded across clinical modalities and anatomical sites:
- Modalities: Computed Tomography (CT), Digital Radiography (DX/CR), Magnetic Resonance Imaging (MR), Positron Emission Tomography (PT), Radiofluoroscopy (RF), Nuclear Medicine (NM), Ultrasound (US), Angiography (XA), and Mammography (MG).
- Anatomical coverage: Chest, brain and head, abdomen, pelvis, spine, musculoskeletal structures, heart, breast, and esophagus.
- Data formats: Native Digital Imaging and Communications in Medicine (DICOM) files, alongside annotations and clinical records in DICOM SR, DICOM SEG, NIfTI, JSON, CSV, and TSV formats.
- Clinical metadata: Demographic variables, laboratory tests, procedures, medications, and disease severity scores (such as mRALE).
- Sequestered benchmarking data: A reserved partition of imaging data is kept sequestered to provide an uncompromised external validation benchmark for AI models, preventing training data contamination during regulatory evaluation.
How to search and retrieve data from MIDRC
Investigators can query and download cohorts through the centralized MIDRC Data Commons portal (data.midrc.org) using the following workflow:
- Access the portal: Navigate to
data.midrc.org. Datasets may be browsed publicly; downloading data manifests and files requires registering a free account and accepting the Data Use Agreement (DUA). - Open the Exploration interface: Click the Exploration tab on the top navigation bar to open the faceted cohort builder, which organizes records into Cases, Annotations, Measurements, Imaging Studies, Radiology Reports, and Data Files.
- Filter by demographics: Within the Cases tab, use the Demographics panel to select study cohorts by sex, age range, or race/ethnicity.
- Specify imaging modalities: In the Imaging Studies tab, apply filters for study_modality (such as CT or DX), LOINC procedure codes, contrast usage, or examined body region.
- Select clinical criteria: Refine cases using the Conditions or COVID Tests panels to isolate specific diagnoses or laboratory outcomes.
- Review annotations and reports: Filter by the Radiology Reports tab for diagnostic text or the Annotations tab for expert segmentations and quantitative severity scores.
- Export the file manifest: Click the manifest or file-manifest button to download a JSON or TSV manifest containing file GUIDs and checksums, or export clinical metadata via data-csv or data-tsv.
- Retrieve files via Gen3 client: Download images locally using the command-line
gen3-clienttool and an API key, or select export-to-workspace to analyze data directly in a Gen3 cloud Jupyter environment.
Strengths and limitations of the platform
Evaluating MIDRC involves balancing its high-quality curation against specific operational parameters:
| Dimension | Key Strengths | Limitations and Considerations |
|---|---|---|
| Curation & Privacy | Strict de-identification pipelines remove Protected Health Information (PHI); professional society governance ensures DICOM integrity. | Inter-institutional variations in acquisition parameters require preprocessing harmonization. |
| Benchmarking | Sequestered validation partition enables unbiased testing of algorithmic generalization. | Sequestered datasets are reserved for formal validation pipelines and cannot be inspected directly. |
| Data Enclave Linkages | Technical links connect imaging records with external enclaves like N3C, All of Us, and BioData Catalyst. | Cross-enclave analyses require navigating governance rules across partner platforms. |
| Access & Licensing | Open and free access under standard academic and commercial DUAs. | Bulk retrieval of large cohorts requires using the Gen3 command-line client rather than in-browser downloads. |
Alternatives and complementary imaging repositories
Researchers frequently utilize MIDRC alongside other major medical imaging archives:
| Repository | Host Institution / Funder | Diagnostic Focus | Distinctive Characteristics |
|---|---|---|---|
| MIDRC | RSNA, ACR, AAPM / NIH NIBIB & ARPA-H | COVID-19, thoracic conditions, oncology, general diagnostic imaging | Centralized Gen3 commons, sequestered benchmarking partition, multi-society clinical curation. |
| The Cancer Imaging Archive (TCIA) | UAMS / National Cancer Institute (NCI) | Cancer radiology and digital pathology | Curated oncology collections linked with clinical trial registries and genomic data in the GDC. |
| NCI Imaging Data Commons (IDC) | NCI Cancer Research Data Commons | Oncology imaging and digital pathology | Cloud-native architecture integrated with Google Cloud Platform, BigQuery SQL, and DICOM-web APIs. |
| Stanford AIMI Datasets | Stanford Center for AI in Medicine & Imaging | Clinical multi-organ imaging, MRI, radiography | Curated challenge datasets optimized for benchmarking and machine learning reproducibility. |
For research teams deploying software or connected hardware alongside imaging tools, security standards are detailed in what is medical device cybersecurity.
Frequently Asked Questions
Is access to MIDRC imaging data free of charge?
Yes. All publicly available imaging collections and metadata on data.midrc.org are free to access after user registration and agreement to the Data Use Agreement.
Can MIDRC data be used in commercial development?
Yes. MIDRC offers both non-commercial academic and commercial Data Use Agreements to support algorithm innovation while safeguarding patient privacy and data rights.
What file formats are available on MIDRC?
Medical images are hosted in standard DICOM format. Annotations, segmentation masks, and clinical tabular data are provided in DICOM SR, DICOM SEG, NIfTI, JSON, CSV, and TSV formats.
How does the MIDRC Data Commons differ from the Biomedical Imaging Hub (BIH)?
The MIDRC Data Commons (data.midrc.org) provides direct downloads of image files hosted in a centralized repository. The Biomedical Imaging Hub (BIH) is a federated indexing platform under the ARPA-H BDF initiative that enables cohort discovery across multiple independent repositories.
How should MIDRC be cited in scientific publications?
Publications utilizing data from MIDRC must cite the resource and acknowledge funding from the National Institute of Biomedical Imaging and Bioengineering (NIBIB) under contract 75N92020D00021 and ARPA-H.
Sources
- Medical Imaging and Data Resource Center (MIDRC), About MIDRC, Governance and Mission, https://www.midrc.org/about-midrc, November 2026.
- Medical Imaging and Data Resource Center (MIDRC), Frequently Asked Questions, https://www.midrc.org/faqs, November 2026.
- Medical Imaging and Data Resource Center (MIDRC), Gen3 Portal Configuration and Schema, https://data.midrc.org/data/config/gitops.json, November 2026.
- National Institute of Biomedical Imaging and Bioengineering (NIBIB), MIDRC Contract and Program Overview, https://www.nibib.nih.gov/, November 2026.
