Ever the visionary, Paul Chang sees AI as an asset to radiologists. As he sees it, “AI and deep learning doesn’t replace us. It frees us to do more valuable work.” The vice chair of radiology informatics at University of Chicago Medicine takes a quick look through the crystal ball at the four stand-out challenges facing radiology with the rise of AI.
- It will take longer to incorporate and consume AI and deep learning than to adopt them. “We’re just beginning to learn how to consume,” he says. This “early stage” offers the opportunity to socialize, change workflow and impact workplace culture.
- We don’t have the data needed to train and validate deep learning. “These systems are incredibly greedy; they need a lot of vetted data because in a lot of ways deep learning is a lazy approach. The clever approach is machine learning like CAD that uses a preconceived model,” he says. “Deep learning gets to that model, but by using a brute force method, it requires you to feed it lots of vetted data.”
- Healthcare industries don’t have the proper IT infrastructure to feed and consume AI and deep learning data. “It’s time now to build up our IT infrastructures, go beyond our PACS-centric perspective and build a new IT infrastructure to consume these systems,” Chang says.
- Weak data infrastructure and ineﬃcient workflow automation. “The real utility of deep learning is the minimally heuristic cases to improve the business of radiology, not diagnosis,” according to Chang. “When you look at other industries—Google, Amazon—the way they use deep learning is not to so much replace the knowledge worker, but to augment and improve the efficiency, to reduce the error, reduce the variability of the workflow.”