Looking back at SIIM19: 3 takeaways on the future of AI and imaging

Regardless of what session attendees were at during this year’s Society for Imaging Informatics in Medicine (SIIM) annual meeting in Colorado, they likely heard about AI in some capacity, and for good reason.

According to Matthew Lungren, MD, MPH, with Stanford’s Center for Artificial Intelligence in Medicine and Imaging (AIMI), the number of published articles investigating AI in radiology have skyrocketed. In 2015, such articles detailing AI thorax and lung applications jumped from about 480 to more than 800 articles published in 2017. And those based on breast-related AI rose from approximately 550 in 2015 to more than 700 in 2017, he explained during the conference.

And while AI wasn’t the only topic discussed, every issue seemed to be tied to the emerging technology in one way or another. At one session, a conversation on burnout quickly turned to AI as a potential remedy (or contributor), to the technology’s ability to spot abdominal aortic aneurysms on CT scans. AI was everywhere.

With that in mind, below are three AI-related takeaways I gathered from SIIM 2019:

1. AI is nothing without data

Algorithms must be trained and evaluated on high-quality, well-annotated data. And before AI can be properly implemented for clinical use it must be trained and tested across various institutions and practices.

The American College of Radiology and RSNA have taken steps to promote structured AI use cases and have enhanced avenues to encourage data sharing across practices, but many presenters at SIIM still felt hospitals are either unwilling to share their imaging data or aren’t sure how.

“For me, the reality is that it’s the wild west out there,” said Judy W. Gichoya, MD, who recently finished an interventional radiology fellowship at Oregon Health & Science University’s Dotter Institute. “We have a lot of data, a lot of open source models, and a lot is still within institutions. I’m not so sure we really understand how to make it accessible.”

Sure the manpower and time it takes to label images cannot be overstated, but the truth of the matter is if healthcare doesn’t want to share its data it doesn’t matter what technology comes along, explained Woojin Kim, MD, CMIO of Nuance Communications, at the conference.

2. Technology isn’t the only hurdle to adopting AI

Of course the technical side of AI in radiology is a challenge, though an ever-shrinking one, but presenters expressed a number of other barriers to adopting AI.

Among them are federal regulations such as HIPPA and the FDA-approval process. The FDA announced they are working on a new framework for regulating AI-based medical devices, but the inherent adaptive learning nature of such devices makes it hard for continual testing.

During another talk, presenters debated the ethics of AI in radiology, with a particular focus on whether or not institutions have the right to sell patient data used to train AI and concerns over HIPPA violations if such data isn’t properly deidentified.

“As healthcare providers we also have ethical responsibilities, we have to follow the law, but we have to put patients first,” said Patricia Balthazar, MD, a radiology resident at Emory University School of Medicine in Atlanta. “We all had to say first, do no harm. When we sell data to third parties things can get tricky and we don’t necessarily know what they’re going to do.”

3. Radiology and imaging informatics can form the future of AI

On a more positive front, a handful of experts expressed their belief that radiology has what it takes to overcome these limitations and meld AI with radiology in a way that benefits both clinicians and patients.

Most were certain that AI will not replace radiologists, but readers who learn to embrace the coming change will no doubt be in a better position than those who deny it.

“Artificial intelligence is the next large technological advancement and it will change radiology and healthcare drastically, said Raymond Wong, MD, radiology resident at the University of North Carolina School of Medicine in Chapel Hill, to the audience. “I think radiology will be elevated by the use of AI tools and those that embrace change, adapt and influence the way AI is used in their practices will fare better than those who do not.”