Partnerships with artificial intelligence startups can help healthcare organizations advance patient care.

During a session at this week’s RSNA annual meeting, on “Artificial Intelligence and Machine Learning in Emergency Radiology”, Dr. Melissa Ann Davis, chief quality officer at Emory Radiology, spoke about how industry partnerships have helped the system move key initiatives forward and “create an ecosystem of healthcare innovation.”

The system is in the first year of the program, having successfully developed term sheets and implemented pilots for two companies, including one that uses infrared detection and AI to better understand patient flow in the interventional suite. It can produce real-time dashboards within some of the suites.

“Ideally we can leverage this information to understand the variation between room and procedure type, to look at optimal scheduling in the future,” Davis said. “This is an example of non-imaging technology that is AI driven. We spend so much of our time focused on imaging technology in radiology, but we also have to understand how non-imaging technologies are also part of the shifting landscape in our field and will probably impact us in a greater way in a sooner time.”

Speaking about building clinically-relevant machine learning systems, Dr. Robert Moreland of the University of Toronto stressed that algorithms need to improve patient care.

“Far too often, we limit the entire use case to solely the detection or classification point,” Moreland said. “Models are not deployed in a vacuum. The applications need to be tested and evaluated on how they perform in the clinical environment.”

Basic machine learning technologies are either run at the device level for rapid detection of an abnormality or injury, in the scheduling system to prioritize studies based on severity, or operates on the PACS as a “second read.”

AI at the scanner level needs to be fast, while on the scheduling level it should integrate into the workflow, Moreland said.

PACS or RIS upgrades should keep AI model development in mind.

“If you don’t get in there at this stage, you’ll be locked out of the ability to deploy them down the line,” Moreland said. “Fortunately, we are developing some very powerful tools to help in this…natural processing, anonymization tools, other DICOM processing files. A good data scientist should be able to help you. In fact, I’m amazed how often we don’t use data scientists at this level. It’s literally their job to be able to analyze, collect and retrieve data.”

Back to HCB News

X-Ray Homepage