CHICAGO — Deep learning and artificial intelligence (AI) were the clear buzzwords of the 2017 RSNA annual meeting and interest was high, with packed, standing room-only crowds for both the lecture and a simulcast, during which several providers shared how computer algorithms were improving their practices.

Dr. Luciano Prevedello, division chief of Medical Imaging Informatics and director of the 3-D and Advanced Visualization Lab at The Ohio State University Wexner Medical Center, spoke about how his facility used AI to help them prioritize imaging studies.

“The idea here is to make our scanners more intelligent, so that by the time we scan, you can submit those images through inferences and then, based on the predictive probabilities, notify the radiologist to go and read those cases sooner, or reshuffle your work list with highest-priority cases,” Prevedello said.

The facility has a special lab for augmented intelligence in imaging, and three water-cooled supercomputers with 15,000 cores each — the typical home computer has up to eight cores — that use open source tools.

Prevedello said the facility is fortunate enough to have the support of leadership for its AI program, and stressed that it is a team effort.

Dr. Curtis Langlotz, who runs a machine learning research laboratory at Stanford University, spoke about how AI can make the job of radiologists easier and improve patient outcomes. He said computer learning should eventually be able to improve image quality and decrease repeated exams.

“Patients who are on the scanner, maybe they’re breathing, moving (and) the images are of suboptimal quality, that might not be noticed until the radiologist looks at the case an hour or two later,” Langlotz said. “It would be much better if the scanner itself could have an embedded algorithm that might alert the technologist that these images are not the quality that you would expect and you may want to repeat them before showing them to the radiologist.”

Langlotz also addressed the hype surrounding AI and the uncertainty of the future for radiologists, noting that the installation of ATMs didn’t negate the need for bank tellers and MR didn’t displace X-ray. And while airplanes are highly reliant on computers, most people would probably not like to fly without a human pilot in the cockpit.

“I’m glad that the pilot had an autopilot to do things that he was not very good at, the monotonous parts or the data integration parts, or the repetitive parts,” Langlotz said. “On the other hand, I’m glad there’s a human there to take the controls when needed and to understand when the autopilot isn’t telling you the information that you need to know, or really shouldn’t be flying the plane. I think that’s going to be our role in radiology. We’re going to get better autopilots to augment what we do and make our lives better and easier.”

Back to HCB News

U.S. Healthcare Homepage