Genetic testing gets most of the attention when it comes to matching cancer patients to specific treatments. But the vast majority of diagnostic information used in selecting a cancer treatment today comes not from those tests, but from under the microscope of the pathologist, who examines tissue biopsies taken during surgery. Stanley Robboy, vice-chair for diagnostic pathology at the Duke University Cancer Center, spoke with Newsweek‘s David H. Freedman about how artificial intelligence (AI) is improving diagnoses—and why some doctors are reluctant to embrace it.
Is precision medicine changing what pathologists do?
Precision medicine has been advancing for a long time. My father was a medical student when penicillin was discovered in the 1920s. For many years that was a single treatment given for infection, regardless of what the infectious organism was. By the time I was in medical school in the 1960s, we had specific antibiotics for specific organisms. For many decades, different cancers were treated with the same drugs.
How is AI helping?
Typically in a cancer case I’ll have to look at 30 to 40 lymph nodes to tell if the cancer has spread. Finding one or two cells can make a real difference in a patient’s prognosis. A single lymph node has about 300,000 cells and I have go through each slide trying to figure out if a single cancerous cell is hiding in there. It’s like looking for the one gray stone among 300,000 white stones on a beach. AI can be terrific at finding that one stone by going through the image in a methodical way, which means the patient is more likely to get the right diagnosis.
Do some pathologists resist AI?
The issue of trust is a difficult one with AI. It’s true, there have been a lot of naysayers, and people who worry that AI will replace them. I don’t ever see turning over the final diagnosis and decision-making to a system. These breakthrough tools will help me, not replace me. They’ll relieve me of some of the most tedious, time-consuming work, freeing me up to focus on the more complex tasks. They’ll look at more of the slide in more detail than I can. And they can give me a second opinion, maybe saving me from making a catastrophic mistake.
What about the impact of AI on the research side?
In my father’s day, an individual physician would do the work that helped determine whether or not a drug worked for certain patients. Ten years ago, three of us would team up to do that work. Today there might be 20 or even 40 people working together on a problem like that. In the future, an AI system will be a critical member of that team, too.