VIDEO: Radiology AI trends at RSNA 2022

He says artificial intelligence (AI) is one of the top keywords readers use when searching through radiology journals and research—and RSNA 2022 clearly showed this correlation of interest on the expo floors and in educational sessions. 

“Everyone wants a piece of AI, this shiny, fancy new technology,” Parekh explains. “But radiologists may now have passed that point where they look at a new algorithm and say maybe its a shiny box and I don't know what to do with it and if I need it. This has now moved to understanding more of what AI is and understanding the potential of AI and where we can go with it to improve clinical care. Radiologists also have become more discerning in what they want.”

Pre-COVID, radiology AI was all about single-point solutions that performed one task. This could be measuring specific anatomy or conditions, automated detection for one type of lung or brain abnormality, or helping solve one piece of radiology workflow bottlenecks. But at RSNA 2022, Parekh observed, radiologists were looking to get more out of the AI in sophisticated solutions that offer fully integrated workflows.

Several radiology vendors developed App Store-like platforms to buy AI apps a few years ago. But this has now evolved into full and seamless integration of these applications into PACS or enterprise imaging system workflows. As more and more AI gains regulatory approval—more than 250 AI algorithms are now cleared by the FDA—more than a few are being packaged together.

“What I have started to see at this show is more than just being a platform that hosts several algorithms across different specialties or clinical use cases,” Parekh tells Radiology Business. “It’s now a case of curating the algorithms and putting them into packages or workflow solutions that address a service line.”

He names as an example a combination of AI algorithms that combine into a package for evaluating CT scans of the brain or lungs to look for multiple types of disease, conditions or trauma. In this way, PACS vendors are now looking at purveyors of best-in-class AI solutions to partner with these companies for full integration into the PACS vendor’s workflow.

PACS, enterprise imaging, workflow management and advanced visualization vendors are also looking for complementing AI solutions that can support the radiologist with things like better organization of DICOM worklists in which critical findings may automatically move the exam to the top of the reading list. Also of interest are things like single-click marking of exams for follow-up. This makes for greater efficiency when tracking incidental findings, taking automated measurements, coordinating tasks like cardiac strain or comparisons with prior exams for tracking changes in tumors. 

“It’s not just the image analysis, it’s also the workload balancing, the fleet management, or how do you get all these resources onto one platform to deliver value to the healthcare system, not just the radiologist,” Parekh says. 

This includes looking at both upstream and downstream workflows and what clinicians need and how AI can be implemented to help facilitate better care by reducing bottlenecks or areas where there is a lot of manual data entry. In acute care this includes several AI vendors to aid acute care teams with alerts, immediate access to imaging and other patient data, and the ability to message everyone at the team before the images are even read by the radiologist.

Parekh says this also includes the trend toward “opportunistic screening” by AI to search for incidental findings in all types of medical imaging that could help improve care by catching diseases before they become more advanced and symptomatic. This includes things like lung screening CT or virtual colonoscopy exams where AI algorithms look for calcium in the arteries to estimate cardiac risk scores.

“This is looking downstream from radiology and saying, OK, we are picking something up opportunistically, but then we don’t want the patient to fall off the radar. You want that to be followed up,” Parekh says. 

AI that can automatically route this incidental finding information to initiate follow-ups with cardiology,  pulmonology or oncology could greatly improve patient care by catching more disease at earlier stages, when it is more treatable and preventable. 

He suggests the present dynamic offers an outstanding business opportunity for health systems to care for and follow-up with these patients. Parekh says the American Medical Association (AMA) now has CPT codes to track usage of technology like this, which is a first step when evaluating if new technologies should receive reimbursement.

Even if AI is not reimbursed, Parekh said healthcare providers need to look at whether AI can help improve patient outcomes and save money in other areas, which could make the investment in AI worthwhile. Beyond opportunistic screening, if AI apps can help save radiologists enough time to be able to read a few more exams each day, there could be business ROI over the long term.