Top Ten Tech Trends 2017: For Imaging Informatics, Focus on Value Puts Emphasis on Analytics, Interoperability | Healthcare Informatics Magazine | Health IT | Information Technology Skip to content Skip to navigation

Top Ten Tech Trends 2017: For Imaging Informatics, Focus on Value Puts Emphasis on Analytics, Interoperability

March 24, 2017
by David Raths
| Reprints
What impact will deep learning have on radiology?

Imaging informatics has traditionally been siloed from other areas of healthcare IT, but many of the exciting developments today involving enterprise imaging strategies, analytics, interoperability standards and deep learning promise to link imaging more closely to the heart of the health system.

The move from an era of volume-based to value-based imaging is already upon us, and that is driving an increasing use of analytics, says Rasu Shrestha, M.D., chief innovation officer at UPMC, the 20-plus-hospital integrated health system based in Pittsburgh, and executive vice president of UPMC Enterprises, its technology development arm. “You can’t push for value-based imaging without having the right sets of measures and analytics in place.” Initial metrics tended to look at efficiency measures such as turnaround time and number of studies read. “In the new world of value-based imaging, it is looking at things like superior outcomes, clinical quality metrics, total cost management, shared savings, and care coordination. These metrics are much more difficult to get at, but here at UPMC we are working to track the right metrics to incentivize the right sorts of behaviors.”

Rasu Shrestha, M.D.

Imaging has to work more closely with other players in the value chain, Shrestha says. In leveraging analytics, UPMC is trying to allow for radiologists to be not just diagnosticians, but also physician consultants. “Their journey doesn’t end with the report going out the door,” he explains. If anything, that is the start of an additional chain of values, where they can be a remarkable contributor to the decision process around the care pathway. The radiology report becomes a living document and then we get to better outcomes down the line.”

The Potential of Deep Learning

IBM’s Watson isn’t the only game in town when it comes to artificial intelligence. Pattern recognition, which is sort of like machine learning, has been used in imaging for quite a while, but it has never been extremely successful, says Bradley Erickson, M.D., Ph.D., professor of radiology and associate chair of research in radiology at the Mayo Clinic, Rochester, Minn. But more recently “deep learning” has shown great promise. “My lab and a few others are starting to get pretty good success at applying deep learning to medical images for making traditional diagnoses of a tumor,” he says. “With traditional machine learning, when we would calculate things like intensity and edges, we had 80 to 85 percent accuracy, but with deep learning, we are now getting better than 90 percent accuracy,” Erickson says. Researchers also are starting to be able to figure out genomic properties of tumors with images, he adds.

Erickson says there are still many unanswered questions about how deep learning algorithms and databases might be used, including how the Food and Drug Administration (FDA) will approve devices like this. “Traditionally, one can understand how an x-ray machine or MRI scanner works. With deep learning, it is much harder to understand how or what property of an image you are looking at in order to make a determination.”

Bradley Erickson, M.D., Ph.D.

At a recent imaging informatics meeting, Erickson participated in a spirited debate about whether computers would eventually replace radiologists. Some people at the meeting said there is no way the FDA would approve the technology’s use until it is better understood. “Interestingly, a company called Arterys got its application approved by the FDA,” Erickson notes. (Arterys says its Cardio DL uses deep learning to automate time-consuming analyses and tasks that are performed manually by clinicians today. It claims the software produces editable automated contours, providing precise and consistent ventricular function in seconds.)

Deep learning companies focused on healthcare are raising a lot of venture capital money, Erickson notes. “I wouldn’t be surprised if there are a number in front of the FDA right now.”

Progress on the Standards Front

One of the hottest topics in healthcare informatics has been the new standard FHIR (Fast Healthcare Interoperability Resources), which uses REST-based application programming interfaces (APIs) to access clinical data such as appointments, orders, and patient information. FHIR also is drawing interest in imaging, as is another imaging-specific set of standards called DICOMweb. “For years we have had the ability for one DICOM device to query another DICOM device,” explains Don Dennison, an imaging informatics consultant who sits on the board of directors of the Society for Imaging Informatics in Medicine (SIIM). “But the fact that DICOMweb promises access to metadata about a study, actual study data, or representations of that study data rendered by a web browser is a very important transition in our community,” he says. DICOMweb, he adds, promises to make retrieving data dramatically more efficient than current DICOM transactions, and like FHIR it uses common web technologies that many programmers are familiar with. 


Get the latest information on Health IT and attend other valuable sessions at this two-day Summit providing healthcare leaders with educational content, insightful debate and dialogue on the future of healthcare and technology.

Learn More