Interest in artificial intelligence (AI) and machine learning at RSNA 2017 seems like it’s unprecedented—but the increased attention is quantifiable. More than 100 sessions delve into the topic at this year’s show in Chicago. Two years ago, less than 10 touched on such concepts.
Paul Chang, MD, a professor and vice chair of radiology informatics at the University of Chicago School of Medicine, urged caution when envisioning the future for the technologies, both in its promise and drawbacks. At “Deep Learning & Machine Intelligence in Radiology,” a session he moderated on Nov. 26 at RSNA 2017 in Chicago, he compared AI to PACS, EMR and other technology hyped as a cure-all to problems faced by radiologists and imaging professionals.
“I’ve seen this before,” he said. “We’ve been on this ride before with disruptive technologies—with PACS, EMRs, Big Data.”
Chang questioned the hype, while still recognizing the possibilities for machine learning in medical imaging applications. Artificial intelligence (AI) has plenty of potential, with deep learning and machine intelligence more specific applications under the AI umbrella.
“We typically as a field tend to too early embrace the hype before the technology is mature enough to really help our patients,” he said. “But ultimately, we can be late adopters once it has become mature.”
Radiologists are expected to view more images and deliver more reports while maximizing value.
Examining what it will take for radiologists to fully realize the potential of AI and machine learning, Chang identified a need for developments in IT infrastructure, which may not be a natural candidate for such investment.
It’s a Catch 22, Chang said, because the potential of advanced applications requires infrastructure improvements. Value in IT infrastructure results from quality, efficiency and safety.
“The danger is complacency,” he said. “The belief that imaging informatics has been solved is a misconception. We need to respond and realize the IT informatics offerings are relatively immature.”
Challenges remain in the validation of advanced systems and the development of more advanced systems.
“Stay calm,” Chang said. “We’ve gone through this many times. … We’ve already redefined ourselves incorporating new technology. Deep learning is not a horrible threat or a savior.”