Circle Cardiovascular Imaging (Circle CVI) now turns a specific data request into a matched, stratified, multi-site cardiovascular dataset in about a week—work that used to take one to twelve months of site-by-site research agreements.

OneMedNet's iRWD™ platform supplies that data under a multi-year subscription, and Circle applies its cardiovascular imaging expertise to train, test, and validate the software inside cvi42.

At a glance

  • About one week from data request to matched cohort.
  • 2,300+ healthcare partner sites across the OneMedNet network.
  • 90M+ patient journeys represented in the network.
  • Cardiac CT and cardiac MR, accompanied by clinical reports and DICOM metadata.
  • Used for algorithm development, testing, and validation, including U.S. FDA submission datasets.

“We needed access to data so we could procure the specific data requirements needed for development, testing, and validation.”

— Karen Arlt, Technical Project Manager, Software Development Core, Circle CVI

The challenge: the right data, not just more data

Cardiovascular AI has to perform across scanner vendors, sites, protocols, and patient populations. Proving that requires datasets built to exact inclusion and exclusion criteria—and those are hard to get.

Beyond large public research cohorts, Circle's data historically arrived through key opinion leader research collaborations: a site shared studies in exchange for research licenses. That model had four recurring problems:

  • Slow agreements. A research agreement took at least a month from first draft to signature, and some sites took six months to a year.
  • Poor fit. Sites shared what they had, not what a project needed. A specific sequence for an FDA validation dataset was hard to source from any one institution.
  • Narrow sources. One site typically meant one scanner manufacturer. A dataset balanced across GE, Siemens, and Philips meant negotiating with three sites, each of which might not have the data.
  • Manual transfer. Data arrived when the site had time, through general file-sharing tools—secure, but not streamlined.

“If we were to wait until a grab-bag mix of different sites provided us the right dataset, it would take a long time to curate the proper data we need for a project.”

— Karen Arlt, Circle CVI

The solution: one request, many sites

OneMedNet contracts with provider sites across its network and queries their archives on Circle's behalf. Circle writes one request; OneMedNet returns one stratified cohort. Its network spans more than 2,300 healthcare partner sites and over 90 million patient journeys.

Three capabilities made the difference for Circle:

  1. Stratification in a single request. Circle can ask for, say, 100 contrast cardiac MR cases balanced across scanner manufacturers and receive them through one channel.
  2. First-party, direct-from-source data. All of OneMedNet's hospital and clinic sources are in the United States, which is especially valuable for Circle's FDA submission datasets.
  3. Complete, documented data. Every delivery includes a data inventory spreadsheet of the DICOM metadata, with pre-negotiated and agreed-to fields—such as disease state or contrast agent—extracted up front. Studies also arrive with their clinical reports, all in the same format and de-identified in the same consistent way regardless of the source facility.

Those clinical reports have grown in value. They give Circle the reason for scan, findings, and impressions alongside the images, which now supports its AI-assisted reporting development and helps interpret DICOM metadata.

“If I ask them for 100 cases of contrast MR stratified across the different vendors, they'll provide that for me in one funnel. That has really helped, especially with our FDA validation datasets.”

— Karen Arlt, Circle CVI

Why a subscription

Circle began with a single scoped project in December 2024 and moved to a subscription in 2025. It is the only real-world data vendor Circle holds on subscription. Circle cites four reasons:

  • Predictable cost. A fixed per-case rate applies to every study, including scarce data that would otherwise carry premium pricing.
  • Flexibility. Circle commits to an annual case volume, and unused allotment rolls into the next year.
  • A working rhythm. Repeated projects have built shared expectations and a steady cadence between the teams.
  • Curation built around Circle's requirements. OneMedNet works with Circle to scope each request—inclusion and exclusion criteria, scanner mix, and the metadata fields to extract—so every cohort is assembled to the project's specification rather than drawn from whatever a site has on hand.

How the collaboration works

Circle defines the cohort, OneMedNet finds and delivers it, and Circle's own quality-control process decides what gets paid for.

  1. Circle defines the cohort. Inclusion and exclusion criteria, date range, scanner manufacturer, and sequence.
  2. OneMedNet queries the network. Partner-site archives are searched and case counts are reported, usually within a week.
  3. Circle confirms the order. For new data types, Circle can request a few sample cases first.
  4. OneMedNet curates and delivers. De-identified DICOM images, a metadata inventory, and clinical reports are delivered through secure SFTP.
  5. Circle re-anonymizes and checks quality. Image analysts check quality and criteria, with duplicate checks across projects.

Circle pays only for cases that pass its own quality control. Accepted cases count toward the annual subscription and carry a perpetual license, reusable across Circle projects. Rejected cases are not billed and are deleted from Circle's systems.

OneMedNet acts as the intermediary: it holds the agreements with provider sites, so Circle does not negotiate site by site. OneMedNet fully de-identifies the data via HIPAA Safe Harbor—Expert Determination is also supported—and harmonizes it as it comes in from many sources. Circle then re-anonymizes every study under its own data-processing standard operating procedure before use. Accepted cases carry a perpetual license: no resale, but no restriction on reuse across Circle projects. Duplicate checks ensure a case is bought only once.

Results

The biggest change is time: Circle now hears back on a data request in about a week instead of waiting one to twelve months for a research agreement.

  • Time to data: query results typically arrive within one week of a scope of work, rather than one month to one year to execute an agreement before site availability.
  • Fit to project: cohorts are built to Circle's inclusion and exclusion criteria rather than whatever a site chose to share.
  • Scanner diversity: cases are stratified across manufacturers in one request rather than usually limited to one manufacturer per site.
  • Sites to coordinate: Circle works with one partner rather than one agreement per site.
  • Documentation: deliveries include a DICOM metadata inventory plus clinical reports.
  • Cost basis: a fixed per-case subscription rate applies, and Circle pays only for accepted cases.

Where the data has gone to work

Since December 2024, OneMedNet data has supported Circle projects across both cardiac CT and cardiac MR. The data has been leveraged for validation of modules requiring regulatory clearance, for research into scarce, high-demand areas of interest, and for future product development.

For routine data, Circle now goes to OneMedNet first rather than negotiating with multiple research sites.

“I know they can run the queries and probably find me this data stratification quicker than me going through four different research sites.”

— Karen Arlt, Circle CVI

What it means for cvi42 users

Validation data drawn from many U.S. sites and multiple scanner manufacturers helps Circle evaluate how its algorithms perform beyond the environment where they were built. Faster access to fit-for-purpose data supports faster development and validation of new cvi42 capabilities.

Lessons for imaging software developers

Circle's experience points to five practices any imaging AI team can adopt with a real-world data partner:

  1. Define the cohort precisely. Inclusion and exclusion criteria, date ranges, and scanner specifications get faster, better-matched results.
  2. Sample before you scale. For an unfamiliar data type, request a few cases before committing to the full cohort.
  3. Keep your own quality-control gate. Re-anonymize, check image quality, and verify every case against your criteria before it counts.
  4. Ask for the clinical context. Reports and metadata inventories become more valuable as your product roadmap grows.
  5. Match the commercial model to your cadence. If you validate continuously, a subscription with a fixed per-case rate and rollover beats pricing each project separately.

Looking ahead

Circle will keep using OneMedNet alongside other sources as it develops and validates new cardiac CT and MR capabilities.

Download the full Circle CVI + OneMedNet case study (PDF)

About Circle CVI

Founded in 2007 and headquartered in Calgary, Canada, Circle Cardiovascular Imaging Inc. develops cvi42, a unified, AI-driven platform for the analysis of cardiovascular MR, CT, and interventional planning studies. cvi42 is used in more than 2,100 hospitals across 90+ countries, supporting millions of cardiac exams every year. Circle CVI is dedicated to advancing cardiovascular care by delivering fast, reproducible, and clinically meaningful imaging insights. Learn more at circlecvi.com.

About OneMedNet

OneMedNet Corporation is revolutionizing Real-World Data through its iRWD™ platform, delivering first-party, direct-from-source, regulatory decision-grade, AI-ready datasets that include de-identified medical imaging alongside comprehensive clinical records. With a network spanning more than 2,300 sites and encompassing over 90 million patient journeys and 270 million studies, OneMedNet serves life sciences companies, medical device manufacturers, AI developers, and other innovators seeking high-quality, compliant healthcare data. The company's platform supports applications ranging from drug development and regulatory submissions to foundational AI model training.