Why RadNet is betting heavily on AI to reshape radiology workflows

He said many AI vendors currently offer point solutions that create operational headaches for healthcare providers because each tool functions independently. RadNet’s goal is to consolidate those capabilities onto a common platform managed centrally by the company. The company can also use its large network of radiologists, imaging centers and hospital partners as a test ground for AI implementation to figure out which ones work well, work out any bugs and incorporate it into a seamless workflows so it is more user-friendly.

“The difficulty many of the developing AI companies have right now is each one requires its own implementation and maintenance,” Berger said. “The goal here is to put them all on one base system and then have RadNet be responsible for maintaining the overall functionality of the tools.”

Berger said the approach is intended to improve operational efficiency while also streamlining the patient experience and supporting radiologists facing increasing workloads.

Building AI around real-world radiology workflow

Berger argued that many traditional imaging IT vendors and OEMs have historically developed software without fully understanding how radiology departments actually function day-to-day.

“The problem that I think most of the OEMs have is they don’t understand the workflow that is required to both utilize their equipment and give the results both to the patient and the referring physicians,” he said.

Because RadNet operates imaging centers directly, Berger said the company has unique insight into workflow bottlenecks, patient anxiety points and operational inefficiencies that can be addressed through automation and AI integration.

He described imaging as a particularly stressful experience for many patients because of fears surrounding potential diagnoses. AI-enabled workflow improvements, he said, can help reduce delays and improve communication throughout the imaging process.

“Trying to make this a more comfortable experience based on our own operational knowledge of what are the kind of touch points and pressure points that we feel will both improve the overall experience for the patient as well as get out a better result on a timely basis,” Berger said.

The company has recruited executives from major healthcare IT vendors to help build out DeepHealth’s platform, but Berger emphasized that RadNet’s operational experience differentiates its strategy from traditional vendor-driven approaches.

“We have the ability to interface between the AI side of the business, the technology and innovation side of it, as well as how that is managed from an operational standpoint,” he said.

Automation aimed at addressing workforce shortages

Berger said automation is increasingly critical as radiology and healthcare systems continue to struggle with staffing shortages and rising imaging volumes.

“There’s almost every facet of the experience that we encounter in dealing with a patient’s experience in one of our centers is something that we’re looking to try to make less manually intensive and more accurate because people make mistakes,” he said.

Examples include AI-assisted scheduling, automated data entry verification, workflow orchestration and remote operational oversight for imaging scanners and technologists.

Berger said relatively simple administrative errors, such as incorrect insurance or patient identification information, can create downstream delays in reporting and reimbursement. Automation could reduce those issues while improving productivity.

“Radiologists get tired,” Berger said. “Things that we can do to make this a more automated function is to the benefit of everybody.”

He added that the timing is critical because imaging demand continues to rise while staffing shortages show little sign of easing.

“The workforce shortage that exists today in healthcare and particularly in imaging is not going to go away anytime soon,” Berger said. “Given the demand that we have, which is at all-time highs here, we need to find more efficient ways to make certain that patients get in quickly and get their results quickly.”

Leveraging scale and data for AI development

A major advantage for RadNet is its massive imaging network and access to large volumes of imaging data that can be used to train and refine AI systems. At the time of RSNA, Berger said RadNet operated approximately 435 imaging centers across nine major markets, and performs roughly 12 million imaging exams annually. About 40% of the centers are joint ventures with hospitals and health systems.

He said those partnerships help shift outpatient imaging away from overcrowded hospitals while allowing imaging to be managed by organizations with specialized outpatient expertise.

The scale of RadNet’s imaging operation also creates what Berger described as one of the industry’s strongest AI development environments.

“Large language models thrive on data and I think our datasets are probably maybe not only the largest, but the best in the world,” Berger said.

He said those data resources have accelerated development timelines for AI companies acquired by RadNet and allowed new products to be deployed more rapidly throughout its imaging network.

Moving toward a unified imaging ecosystem

“I think the future for imaging will best be delivered if we don’t worry about where a patient gets their scan, but who’s available to read it with the most qualifications at a given moment,” Berger said.

He noted that imaging now occurs across a broad range of settings, including hospitals, outpatient imaging centers, physician offices, urgent care centers and emergency departments. AI-enabled workflow systems, he said, could help unify disconnected environments.

Hospital systems are increasingly interested in those capabilities as they face growing radiologist shortages and mounting imaging demand.

“They’re now also being confronted with huge radiologist shortages and are looking for alternatives where they can augment the need for the reading component, which is being massively challenged,” Berger said.