Facing a radiologist shortage that shows no sign of easing, twelve of the largest health systems in the United States announced on August 11, 2026 that they are banding together with AI vendor Aidoc to form the Diagnostic AI Consortium, a coalition built to test, validate, and share AI-enabled diagnostic workflows across radically different hospital environments. The founding members, Advocate Health, Cedars-Sinai Health System, Hartford Healthcare, Houston Methodist, Mercy, Mount Sinai Health System, Northwell Health, Northwestern Medicine, Sutter Health, University of Florida Health, University Hospitals of Cleveland, and WellSpan Health, together serve nearly 20 million patients a year, giving the group unusual scale for what is effectively a shared experiment in how far AI can be trusted to touch diagnosis.
A shortage that keeps compounding
The consortium’s own framing leans heavily on numbers that describe a system under strain. Imaging interpretation turnaround times roughly doubled between 2014 and 2023, even as scan volume kept climbing, and radiologists have been leaving the field at a rate about 50% higher than before 2020. Workforce projections cited by the group suggest the shortage will persist through 2055 without some form of intervention, a timeline long enough that hospital executives are no longer treating AI adoption as optional experimentation but as a structural necessity for keeping diagnostic turnaround times from getting worse.
What Aidoc actually brings to the table
Aidoc’s AI already operates inside nearly 2,000 hospitals worldwide and analyzes roughly 60 million patient cases annually, giving it one of the largest deployed footprints of any clinical AI vendor. For the consortium, Aidoc is contributing two pieces of infrastructure: its CARE foundation model, trained to flag abnormalities across a wide range of imaging studies rather than a single narrow use case, and aiOS, an enterprise operating system meant to handle deployment, workflow integration, and continuous monitoring after a model goes live inside a hospital’s systems. That monitoring layer matters because a persistent criticism of hospital AI has been that tools get validated once before launch and then rarely re-checked as patient populations, scanners, or protocols drift over time.
Two guiding principles, and a built-in tension
Aidoc CEO Elad Walach described the effort’s philosophy directly: “This Consortium is built on the belief that it can be done the right way, guided by two principles: iteratively and together.” The iterative piece reflects an acknowledgment that no member hospital expects to get its AI workflows right on the first try. The together piece is the more unusual bet, since the twelve systems are agreeing to measure outcomes using shared methodology and then publish which practices worked, effectively asking direct competitors in some regional markets to pool their diagnostic AI learnings rather than treat them as proprietary advantage.
A voice from inside the consortium
Dr. Leonardo Kayat Bittencourt of University Hospitals, one of the founding systems, offered a more cautious framing than the press release’s headline numbers, arguing that AI in diagnosis “will only reach its full potential through shared expertise and accountability.” That comment points to the harder, less quantifiable part of the initiative: accountability structures for when an AI-assisted read goes wrong, and shared expertise for less-resourced systems that cannot independently build the validation infrastructure that a system like Cedars-Sinai or Northwestern Medicine can afford on its own.
The skeptical view: adoption ahead of governance
Not everyone in health-system leadership is convinced hospitals are ready to scale AI diagnostic tools this fast. Roughly 80% of U.S. hospitals now report using AI in at least one clinical or operational function, according to industry surveys, but experts at institutions like UC Irvine Health have separately warned that many provider organizations do not have full visibility into where every AI tool is already running inside their own systems, let alone a governance structure to track outcomes consistently. That gap between deployment speed and oversight capacity is precisely the vulnerability the Diagnostic AI Consortium says it wants to close, but it also means the consortium is, in effect, racing to build guardrails around a technology already embedded in day-to-day clinical work rather than pausing to build the guardrails first.
What comes next
The consortium’s stated plan is straightforward in structure if ambitious in execution: member systems will deploy AI-enabled diagnostic workflows using Aidoc’s CARE model and aiOS platform, measure impact using agreed-upon metrics, and then share which practices succeeded so that all twelve systems, and eventually the broader industry, can adopt them. If it works, the model could offer smaller and rural hospital systems, the ones least able to run their own AI validation programs, a template they can copy without having to independently prove out every workflow. If it stalls, the reasons will likely be familiar ones: differing electronic health record systems, inconsistent data quality across sites, and the slow, case-by-case work of building clinician trust in flagged findings. Given that the imaging backlog and radiologist shortage numbers underlying the announcement are structural rather than cyclical, the pressure to make some version of this work is unlikely to ease anytime soon.
