Artificial intelligence models help diagnose cancer

Mention artificial intelligence (AI) at a social gathering and you’re likely to encounter strong feelings—both positive and negative.

For Justin Hwang, however, AI isn’t just conversation fodder. It’s his partner in a mission to lengthen and save lives.

Hwang, an assistant professor of medicine in the University of Minnesota Medical School’s Division of Hematology, Oncology, and Transplantation, is deploying AI-powered algorithms to identify which cancer therapies might work best for individual people.

Hwang and his colleagues are developing and using these algorithms to plumb enormous data sets, finding meaningful patterns that can inform cancer treatment decisions. The goal is to save precious time by weeding out therapies unlikely to work for an individual and pointing care teams toward treatments that should.

The approach involves both zooming in and zooming out: zooming in to examine an individual person’s genetic profile, and zooming way out—to a population level—to find patterns among thousands, even millions, of other people with similar genetic markers. 

“It’s almost like we can see how each individual fits into this massive population based on their genomics,” Hwang says. “Is there a pattern related to good outcomes? Bad ones? To drug sensitivity or insensitivity? Why are certain people more responsive [to some treatments] while others are less responsive? We’re essentially using genomics to assign a person’s trajectory.”

Making sense of the search

The exploding availability of DNA, mRNA, patient data from health records, diagnostic images, and other digital sources is key for researchers like Hwang and his colleagues.

Having that data is one thing; making sense of it is something else entirely.

With support from philanthropy, including the Randy Shaver Cancer Research and Community Fund, Hwang and his team are developing machine learning algorithms—a type of AI that allows computers to learn from data sets—to uncover meaning from previously inscrutable mounds of information. Hwang and his colleagues partner with healthcare analytics companies that provide the deidentified data.

“We’re [part of] an oncology alliance with more than 100 cancer centers, and they all use the same test—the data is collected uniformly,” Hwang explains. “We’re sort of the translator, helping [clinicians] say, ‘OK, what does this mean?’”

Hwang shares an example of the type of question his team is out to address: Will a person’s prostate cancer spread to other areas of the body or not? With the help of his machine learning algorithms, he’s found that most people—some 90 percent—will not develop metastasis.

“There’s a huge population that never has to worry; many patients will be perfectly fine with normal treatment,” Hwang says.

Hwang’s lab is making it easier to identify people most at risk of the worst outcomes. So rather than wasting months or years on ineffective treatments, they can focus on therapies more likely to work.

Conversely, people who are likely to have positive outcomes can avoid unnecessary overtreatment—and the expense, side effects, and lost time that accompany it.

Actionable information

Until now, such predictions have “really been a shot in the dark,” Hwang says—but genomic data is changing that.

He says the University of Minnesota is the ideal home for his work, especially due to the collaboration with oncologists in clinical research and trials.

“I’m not sure this would be doable outside of Minnesota. The University is such a collaborative community—and here, patients come to the same clinic for decades,” so there’s rich longitudinal data to mine, Hwang explains. “Over 15 years, we knew who developed metastasis. We just went back and assessed the 20,000 genes, and we were able to say, ‘Oh, here’s where we predict the patients that would get metastasis,’ and we did so with 80 to 90 percent accuracy!”

Using machine learning to glean actionable knowledge from mountains of data is just the first step in Hwang’s work. The next phase involves affirming his team’s findings in clinical studies.

Thus far, Hwang’s research has mostly focused on prostate, breast, ovarian, and pancreatic cancers. But he says one of the most exciting elements of his lab’s work is its seemingly limitless applicability.

“What about hereditary diseases? Alzheimer’s? Psychological disorders? I’m interested in any space that can use genomics to look at outcomes,” he says. “I think this is just the tip of the iceberg.”

Make a gift to support the Masonic Cancer Center's next-generation cancer research.