Twelve large US health systems have formed a consortium to develop shared methods for evaluating, implementing and governing diagnostic artificial intelligence. Together, the participating organisations care for nearly 20 million patients each year. That scale should allow them to assess whether AI-supported diagnostic workflows remain safe and effective across different hospitals, patient populations, imaging equipment and clinical settings.
The Diagnostic AI Consortium includes 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. Clinical AI company Aidoc will provide the technical infrastructure. The first findings are expected to be shared in 2027.
The collaboration marks a shift away from evaluating individual algorithms in isolation. Its central question is broader and more relevant to daily care: what happens when diagnostic AI becomes part of an entire clinical workflow?
Shared evidence across diverse health systems
Diagnostic AI is already used to identify or prioritise suspected abnormalities in medical images and other clinical data. Yet strong performance in a controlled study does not guarantee that a system will work equally well in another hospital. Differences in scanners, patient populations, disease prevalence, data quality and working practices can all affect performance.
This is one of the problems the consortium intends to address. The participating health systems will jointly design AI-supported workflows and study their effect on diagnostic safety, quality and speed. Among other things, they want to determine whether AI can help identify urgent cases earlier, accelerate interpretation and ensure that critical findings reach the appropriate clinical teams sooner.
The partners also plan to turn successful approaches into shared implementation and governance practices that could be adopted by healthcare organisations outside the consortium. This includes questions that become increasingly important once AI moves beyond a local pilot: how should a system be validated before deployment, who remains responsible for its performance and how should hospitals detect bias or declining accuracy after implementation?
“No single centre, however large or reputable, can capture the diversity of imaging, disease and clinical context needed to build and deploy models that generalise safely,” said Leonardo Kayat Bittencourt, Vice Chair of Innovation at University Hospitals, in the official announcement.
Diagnostic capacity under pressure
The initiative is being launched at a time when diagnostic services in the United States are struggling to keep pace with demand. An analysis by the Harvey L. Neiman Health Policy Institute found that the average interval between an outpatient imaging examination and its interpretation increased by 27 percent between 2023 and 2024.
The analysis covered 2.9 million imaging examinations among Medicare beneficiaries. Over the full period from 2014 to 2024, interpretation turnaround times rose by 177 percent, with 92 percent of that increase occurring in the final three years. The trend differed substantially by imaging modality: between 2023 and 2024, turnaround times increased by 49.1 percent for ultrasound, 35.4 percent for radiography and fluoroscopy, 11.6 percent for CT and 4.5 percent for MRI. The Neiman Institute published the results in August.
These figures do not prove that AI will resolve the underlying capacity problem. They do, however, explain why health systems are looking beyond tools that merely detect abnormalities. Prioritising worklists, routing findings and connecting results to subsequent clinical action may ultimately be just as important as the performance of the detection algorithm itself.
That wider view is reflected in the consortium’s plans. Its members want to follow the diagnostic pathway from the moment an examination is completed through interpretation, communication and clinical follow-up. The ambition is not simply to produce another AI alert, but to establish whether technology can shorten the route from a clinically relevant finding to appropriate action.
Aidoc’s central role in the consortium
Aidoc is more than a technical supplier to the consortium. Founded in 2016, the company initially developed AI that alerted radiologists to suspected acute abnormalities in medical images. It has since expanded into an enterprise platform through which hospitals can deploy, integrate and monitor multiple clinical AI applications. Aidoc says its technology is now used in nearly 2,000 hospitals and analyses more than 60 million patient cases annually.
The company has also attracted substantial investment. In April 2026, Aidoc raised $150 million in a Series E funding round led by Goldman Sachs Alternatives, bringing its total funding to more than $500 million. The capital is intended partly to expand CARE, its clinical foundation model, and aiOS, the platform that will underpin the consortium’s work.
That common technological foundation should make it easier for twelve health systems to compare implementation experiences and monitor AI-supported workflows at scale. At the same time, this is not a vendor-neutral evaluation of competing diagnostic AI systems. Its wider value will therefore depend on whether the consortium publishes methods, outcome measures and governance lessons that healthcare organisations can apply independently of Aidoc’s products.
For now, the consortium represents an intention rather than evidence that diagnostic AI improves patient outcomes or relieves workforce pressure. Its significance lies in the decision by twelve health systems to evaluate the technology collectively and under shared standards, instead of allowing each hospital to repeat the same validation and implementation process on its own.
If the partners publish transparent results across diverse clinical environments, the initiative could provide a useful model for healthcare systems elsewhere. The decisive step will come in 2027, when the first results are expected to show whether cooperation at this scale can move diagnostic AI from technical promise to demonstrable clinical value.
