AI takeaways from RSNA 2018

Without question, artificial intelligence (AI) is the hottest topic at the RSNA 2018 Annual Meeting. It’s not even close. It all started with the opening session on Sunday, November 25, when RSNA President Vijay Rao, MD, spoke at length on the topic of AI, saying it is something radiologists should embrace rather than fear.

To Rao, AI is not a threat to the profession, but a tool that can empower radiologists to spend more time on important initiatives that can benefit both physicians and patients. Brian Bialecki, from the American College of Radiology, echoed these sentiments during our Facebook Live broadcast on Tuesday, November 27, when he said that he really views AI more as “assisted intelligence” for radiologists rather than “artificial intelligence.”

“Most AI algorithms are meant to assist the radiologist in their findings and their everyday workflows,” he said. “For example, a vendor may create an algorithm that is great at detecting lung nodules, but there are hundreds of thousands of other abnormalities that may be present on that scan that the algorithm doesn’t account for. It’s up to a radiologist to determine whether or not those conditions exist. You will always need the human element of the radiologist to act as a gatekeeper who ensures you consider all possible interpretations of an image.”

AI could be a radiologist’s best friend

You can see the possibilities of AI in the medical imaging profession throughout the RSNA educational sessions and the Exhibition Hall, particularly the Machine Learning Showcase. There’s undoubtedly a lot of hype, but there are also several use cases that show great promise.

For example, AI is proving effective at identifying common cancers and cardiovascular abnormalities, detecting fractures and other musculoskeletal injuries and aiding in the diagnosis of neurological diseases and thoracic conditions. In each of these cases, workflows are altered to improve detection and diagnosis of potentially fatal diseases while improving the productivity (and potentially the accuracy) of radiologists and pathologists – allowing them to ultimately deliver improved service to their physician partners and better care to patients.

Research presented by Enhao Gong, PhD, on Monday, November 26, at RSNA 2018 provided another interesting use case for AI. Gong and his colleagues trained a deep learning algorithm to approximate the effect of full contrast-enhanced images on MRI images where a low dose of contrast or no contrast was used. This breakthrough could potentially reduce the amount of gadolinium patients are exposed to throughout their lifetimes.

Imaging infrastructure changes needed to take full advantage of AI

While the benefits of AI are well within reach for medical imaging professionals, there is much the community needs to do to take full advantage of the technology. One initiative Rao suggests is for radiologists to reformat and rebrand their reading rooms into “digital diagnostic data hubs.”

The purpose of the digital diagnostic data hub is for clinical teams to get together to virtually collaborate on patient care using tools like video conferencing and universal viewing. In this environment, radiologists could rely on AI to aggregate the current imaging findings with those of prior studies from other modalities. They can also gather health history, lab results, biopsy and pathology findings, patient demographics, genomics and more.

By using AI to bring everything into one central location, specialists can make a diagnosis they may have missed if a single exam were considered in isolation.

Bialecki further explained how implementing technologies such as a vendor neutral archive (VNA) can be instrumental in supporting AI (and by extension the concept of a digital diagnostic hub) in his Facebook Live broadcast.

“When it comes to getting the most out of AI, you can’t just depend on the AI algorithms of your chosen PACS vendor,” he said. “You’re going to want to leverage whatever technology provides the best algorithm for a particular finding, regardless of vendor. You’re also going to need the flexibility to distribute, gather and correlate images from a variety of systems and modalities – not just radiology images from PACS. Implementing a VNA in front of your PACS can give you this flexibility and it’s the kind of infrastructure radiologists need to start thinking about.”

Ready to learn more? Watch the video above!

Ken Congdon has expertise in the healthcare technology industry and has been a contributor to the Hyland blog.
Ken Congdon

Ken Congdon

Ken Congdon has expertise in the healthcare technology industry and has been a contributor to the Hyland blog.

... read more about: Ken Congdon