Radiology AI vendors shift focus to workflow integration and enterprise value

AI platforms evolving beyond marketplaces

Earlier expectations that vendor-neutral AI marketplaces, similar to an App Store, would become the dominant distribution model have not fully materialized. Ahmed said the problem is that vendor-neutral environments can make it difficult for individual applications to stand out.

“When you have a vendor neutral platform, there's promise of scalability. Yet if your application is hidden in a vendor neutral swamp of applications, how do you benefit? How do you get out there? How do you market it?” Ahmed said.

There was previously a big focus among AI vendors on specific clinical use cases to identify specific findings. But that has changed to a new trend where AI platforms are increasingly shifting toward analytics, servicing capabilities and deeper workflow orchestration within imaging IT systems.

“Workflow is something that always gets talked about,” Ahmed said. “But what really is seamless integration? It's aligning with the needs of imaging IT and PACS.”

He described imaging IT infrastructure as the “backbone of this entire ecosystem,” noting that meaningful AI integration depends on compatibility with existing systems and workflows.

Expansion beyond radiology departments

Another major trend highlighted at RSNA was the shift toward AI applications designed to support care coordination across hospital systems rather than focusing solely on radiology productivity.

“Radiology AI shouldn't just be purely radiology,” Ahmed said. “It needs to align with the needs of the more holistic hospital networks.”

He pointed to incidental findings as an example of where AI could improve systemwide care coordination. For instance, cardiology-related findings on breast imaging may currently go unaddressed if they are not properly communicated to downstream specialists.

“A growing trend that we're seeing is, for instance, breast mammograms, you can get an incidental finding for cardiology,” Ahmed said. “Now, that incidental finding is found, but it's not alerting the cardiologist downstream and that patient has gone home and almost forgotten about it.”

AI systems that automatically notify the appropriate care team could improve patient follow-up and create additional value for health systems.

“It needs to alert the right person,” Ahmed said. “If possible, keep the patient in-house, keep them in the hospital, get that second scan done, get them seen to.”

Oncology and enterprise care models gain momentum

Care team coordination using AI for acute use cases like stroke, pulmonary embolism and aortic aneurysm has expanded to now include oncology workflows and multidisciplinary care coordination, reflecting global trends in cancer screening and treatment.

“There’s obviously a growing trend across the world around oncology and cancer screening,” Ahmed said. “It's where the need really lies.”

By targeting oncologists and other specialists, AI vendors are positioning their tools as enterprise solutions rather than department-specific software. That has broader appeal to healthcare systems.

“When you're bringing in multiple different personas, suddenly your AI is no longer a boxed-off product in one department,” he said. “It's a product for the entire network.”

Consolidation and new AI operating system strategies

Ahmed also highlighted a growing trend among companies building comprehensive AI suites through acquisitions and partnerships.

One example discussed at RSNA was the strategy of DeepHealth, an AI division created by RadNet, one of the largest radiology providers in the U.S. The business plan its to develop AI algorithms that can help improve RadNet's workflows and patient outcomes. The company can test algorithms in the real world on a large scale. It then can take the ones that work well when vetted and offer them as a products to health systems and other radiology practices.

“DeepHealth's model has worked really well for them,” Ahmed said, noting the company has been among the most active acquirers in the AI imaging space.

RadNet has expanded its breast imaging portfolio through acquisitions such as iCAD, while also targeting lung, prostate and thyroid screening applications. Ahmed said this reflects a broader industry shift toward “operating system” strategies that bundle multiple applications into unified platforms.

“We're seeing companies building comprehensive suites of applications, partnering or acquiring to fill out that portfolio,” he said.

New buzzword is agentic AI

Each RSNA meeting tends to produce a new AI buzzword, Ahmed said. In previous years it was generative AI, while in 2025 the focus shifted toward so-called “agentic AI.” This refers to systems capable of autonomously coordinating multiple tasks across workflows, although Ahmed cautioned that the concept remains largely aspirational for now.

Revenue concentrated in a few applications

Despite the rapid growth of AI in medical imaging, commercial success remains concentrated in only a handful of clinical applications.

According to Signify Research data, the global medical imaging AI market generated about $749 million in revenue in 2024.

“The issue that we are seeing today is that commercial traction is mainly in three or four main areas that make up today's revenue,” Ahmed said.

Among the most successful applications are stroke triage systems now offered by several vendors, and cardiac imaging tools such as fractional flow reserve CT (FFR-CT), which was initially developed by HeartFlow. CT coronary plaque analysis is another, initially spearheaded by Cleerly. These areas have gained traction partly because they align with urgent clinical workflows and, in some cases, reimbursement pathways. Both FFR-CT and plaque analysis now have their own category 1 CPT codes for payments, which is a rarity in the AI space, but a tribute to the investment in studies and clinical data these vendors made.

ROI remains a key challenge

While more than 1,000 radiology AI algorithms have received clearance from the U.S. Food and Drug Administration, only a small number currently have reimbursement through Category 1 CPT codes.

That reality continues to challenge the business case for many AI products.

“It has to come down to saving time across the hospital networks,” Ahmed said.

He added that enterprise-level benefits such as care coordination, incidental finding detection and operational efficiency may ultimately drive adoption more than reimbursement alone.

“When that incidental finding can be found and the right person is alerted, then this is what attracts that wider hospital need for AI,” Ahmed said.

For many health systems, he said, the future success of AI will depend on demonstrating measurable value beyond the radiology department.

“As currently, the ROI just doesn't make sense,” Ahmed said. “It can seem much more favorable if you're benefiting the entire healthcare system and not just one department.” 

Will radiologists be replaced by AI? Read more in the article CEO of America’s largest public hospital system says he’s ready to replace radiologists with AI.

Will radiology technologists be replaced by AI? Read more in the article Nvidia sees major shift in radiology to AI agents and new autonomous imaging systems.