Radiology on cutting edge of clinical decision support

Sessions at the RSNA conference are showcasing cutting-edge applications of clinical decision support -- a key meaningful use requirement -- for medical imaging studies.

CHICAGO -- As a clinical specialty, radiology might be behind the curve with regard to chasing federal health IT incentive payments for implementing electronic health records (EHRs). Nevertheless, there's one meaningful use requirement that holds great potential for radiologists -- clinical decision support.

The reason is that radiologists remain "walled off" from the rest of the health care IT ecosystem, despite all the heavy technology at their disposal for creating images, said Sandy Napel, professor of radiology at Stanford University and co-director of Information Sciences in Imaging at Stanford. Clinical decision support tools are one way to break down those walls and integrate outside data with patient cases, he noted in a session at the Radiological Society of North America (RSNA) 96th Scientific Assembly and Annual Meeting.

Radiologists typically don't have access to EHR systems. They have to use computers separate from their imaging workstations to access the Internet or even to email physicians or colleagues. Their images tend to live on picture archiving and communication systems (PACS) in giant data tanks, without much metadata or organization.

"If you think your digital library at home is overcrowded with snapshots, just look at your PACS system," said Napel, who demonstrated image-recognition technology being developed that can summon studies similar to the one a radiologist is reading. "What's such a crime is that all of these images are just sitting there -- and there's no way to find them. You might be sitting down, looking at a case, saying, 'I know I saw one that looked like this about a year ago, I'd love to see it again, and moreover, I'd love to see what the outcome is, what drug worked,' but there's no way to find the image."

If you think your digital library at home is overcrowded with snapshots, just look at your PACS system.

Sandy Napel, professor of radiology, Stanford University

In his demo, Napel showed liver studies and described how his team's software could scour image archives to find similar abnormalities according to their shape and texture and their borders' smoothness. The images in his research project contained both standardized metadata keyed in by radiologists and computer-generated metadata about the image.

Napel and his co-presenters, Dr. Charles Kahn of the Medical College of Wisconsin, and Dr. Elizabeth Burnside of the University of Wisconsin School of Medicine, agreed that smart computer searches would extend the ability of radiologists to find comparables, which are a key tool in diagnosing patients. They showed radiology software being developed that's based on such artificial intelligence schemes as Bayesian filters and case-based reasoning.

The latter closely resembles the human thought process -- deriving the next decision based on what a person has seen before, Kahn said. Computer applications can "think" similarly, but they never forget and have more processing power to consider more possible matches, he said. The downside is that computers can't explain how their artificial intelligence arrived at their conclusions, so the decision process cannot be analyzed or learned from.

"We are living in an era of information overload," Burnside said. For the radiologist, there are far too many pieces of data to take into consideration -- for example, the cumulative knowledge contained in the patient's EHR, emerging research catalogued in scholarly journals, and weekly updates to genetic cancer risk -- all of which potentially can carry weight in a diagnosis. "Human decision making involves heuristics that often [don't] scale up to this huge amount of data."

Ultimately these tools will be more effective when they're plugged into national, Web-enabled databases of radiological images, and when they're integrated with EHR systems. Napel sees a lot of potential for improving health care with these clinical decision support tools. When they're tuned properly to return accurate results quickly, they could give radiologists the power to order follow-up tests before a patient leaves an appointment, or even to recommend drug treatments based on imaging results compared to a national population of similar cases. Adding imaging data to the Gail Model breast cancer risk assessment tool, for example, could lead to better, more accurate mammography results and more effective treatments, Burnside said.

All three experts emphasized that, although these imaging decision support systems, in limited testing, often could make a diagnosis as accurately as their human counterparts, their real upside is supporting -- not replacing -- humans. A radiologist using a clinical decision support system can identify correctly what a particular image shows more often than can he or she alone -- or the system alone. Alternatively, using such a system could help a novice radiologist become an expert diagnostician.

It all adds up to meaningful use of health IT, Kahn said. That is, radiologists are practicing what the Office of the National Coordinator for Health Information Technology, or ONC, is preaching -- improving health care quality by applying technology -- in a way that is practical for the specialty.

"These are meaningful uses of informatics," Kahn said. "That's what meaningful use was meant to be -- using technology to demonstrate improved outcomes."

Let us know what you think about the story; email Don Fluckinger, Features Writer.

Dig deeper on Clinical decision support systems

Pro+

Features

Enjoy the benefits of Pro+ membership, learn more and join.

-ADS BY GOOGLE

SearchCompliance

SearchCIO

SearchCloudComputing

SearchMobileComputing

SearchSecurity

SearchStorage

Close