ARPA-H backs a national marketplace for hospital scans and pathology slides

Rafid Fadul announced Harbor Health Cloud, a project to make imaging data from different institutions usable together and help hospitals analyze their own collections.

Author
HDM Staff
Save
Share
Matthew Herper and Rafid Fadul seated in conversation on the Main Stage.
STAT’s Matthew Herper interviews ARPA-H Chief Medical Officer Rafid Fadul. Photo credit: HLTH USA 2026.

    The Advanced Research Projects Agency for Health is backing a marketplace for medical images and pathology slides, aiming to let researchers study cases scattered across hospitals. Dr. Rafid Fadul, the agency’s chief medical officer, announced Harbor Health Cloud on Nov. 16 during a discussion with STAT senior writer and editorial director of events Matthew Herper, who moderated the Las Vegas session. The project is led by a team that includes Boston Children’s Hospital and the University of Utah.

    Fadul raised a common problem for researchers looking at uncommon conditions. A single hospital may have only a few relevant cases, and other hospitals keep the images and biopsy slides from their cases in separate repositories. “It’s hard to draw insights in aggregate,” he said. The new award aims to make enough of those images usable together for developers to train medical imaging tools and for researchers to investigate questions that small local collections cannot answer.

    What a hospital might contribute and get back

    The agency’s award description says the project will give developers access to deidentified radiology and pathology images. Fadul described a marketplace where a hospital could contribute images it already pays to store and use tools to examine its own collection. His example was a rural hospital that might have useful material but lack the staff or software to study it at scale.

    The benefits for a contributing hospital will depend on how the marketplace works. Fadul spoke about potential financial upside and research value. The session did not set out fees, participation terms, which organizations could request access or how they would be approved. Those details affect whether hospitals with limited staff can contribute and whether the material available for analysis reflects practice outside major research centers.

    Fadul illustrated the clinical question with acoustic neuromas, also called vestibular schwannomas. A clinician may encounter them infrequently, limiting what can be learned from cases at one institution. Finding more relevant scans across hospitals could help researchers ask which imaging features are associated with what happens to patients. Answering that kind of question would also require reliable descriptions of how the scans were taken and what happened afterward. The award and Fadul’s announcement set out the platform’s purpose. They supplied no evidence yet that it had assembled enough comparable cases to answer a specific clinical question.

    The same distinction applies to developers training an imaging tool. A large collection is useful only if it contains examples relevant to the settings where the tool will be used. Researchers need to know whether an image comes from a routine exam, a referral case or a specific scanner and protocol. Variation across hospitals may be an advantage for testing how a tool performs, provided the people running the study can tell what accounts for that variation.

    Bringing more sites into clinical trials

    Herper also asked what would make U.S. trials more competitive when drug developers can recruit elsewhere. Fadul argued that sponsors need a reason to choose domestic sites and that one way to give them a reason is to make trials easier for more hospitals and clinics to run. He described Texas as a starting point for STACK, an initiative announced in September to help activate trial sites. In his account, recruiting outside major academic centers could connect sponsors with a wider group of patients and reduce the distance some participants have to travel.

    Matthew Herper and Rafid Fadul in conversation on the HLTH USA main stage.
    Matthew Herper and Rafid Fadul on the Main Stage in Las Vegas. Photo credit: Health Data Management.

    Opening a trial at a rural site still requires staff to identify patients, explain the study, document participation and carry out follow-up. Fadul said smaller sites can see research as a cost they cannot readily absorb. A network might lower the burden of starting a study, but participation will also turn on whether the site can cover the work after enrollment begins. For sponsors, the question is whether a site can keep the trial running and produce complete data throughout the study.

    Fadul said trial teams often have to “rebuild the infrastructure of a trial every single time.” He described COMMONS as part of an effort to reuse governance and review arrangements. ARPA-H’s description emphasizes a consent-governed research resource that brings data into a common format. Both problems matter when organizations work together repeatedly. They need a way to decide who can use data and a way to interpret it consistently after access has been granted.

    The related SURPASS program is seeking ways to combine adaptive study designs, faster analysis and computer simulation. Fadul discussed the possibility of allocating more trial participants to a treatment arm if investigators have sufficiently reliable information for comparison. Whether that works depends on the disease, how well the comparison represents patients in the trial, and what the study needs to establish. The discussion offered no results showing that one enrollment ratio, design or timetable would work across treatments.

    Testing a clinical AI system after deployment

    Herper asked about ADVOCATE, an agency program developing AI systems that could support cardiovascular care, including heart failure management. Fadul described a separate layer to supervise the patient-facing software as data and clinical practice change. His aim was for the tool to work on “day one and on day one thousand.” The question for a monitoring system is how quickly it can detect a change in performance and what a clinician or operator can do once it is detected.

    ARPA-H announced initial ADVOCATE awards in September. The program is working toward systems that could earn authorization from the Food and Drug Administration. The discussion did not report that a system had already received authorization or demonstrated an improvement in patient outcomes. Herper’s questions about trials and clinical software pointed to the same issue at different points in development. Getting a tool into more settings creates an opportunity to test it under more varied conditions, and that opportunity matters only if the resulting evidence can be interpreted.

    Harbor’s first test is more immediate. It has to bring in useful image collections from different institutions and give researchers a workable way to study them under agreed terms. If it can do that, a hospital’s archive could contribute to a research question that its own physicians see too infrequently to answer on their own. The announcement identified the organizations building that capability and the pool of information they want to make available. The next evidence will come from the hospitals and researchers that can actually use it.