Diagnosing Skin Cancer With Google Images
Man vs. Machine examines the ways in which technology promises to simplify our lives—or, for some of us, threatens to upend them.
An artificial intelligence algorithm using Google Inc.’s image search has proved capable of diagnosing malignant skin lesions as accurately as a group of 21 certified dermatologists.
The developer, who has no background in dermatology, envisions a future in which everyone can screen themselves for skin cancer with a phone.
- Brett Kuprel
- Age: 26
- Electrical engineering Ph.D. student at Stanford
- Location: Palo Alto
Kuprel began tinkering with lesion diagnostics as a way to refine a cancer-detecting neural net, an AI designed to mimic the brain.
Dermatologist Rob Novoa, who co-authored a paper on Kuprel’s AI in the journal Nature, says he hopes a diagnostic bot can speed the classification part of his clinical work, so he can see more patients and spend more time with them.
A skin-scanning AI might be able to reduce the number of surgeries patients need, but biopsies remain an extremely reliable means of detection—and revenue source—for hospitals, says Adam Friedman, a dermatology professor at George Washington University.
In general, doctors, nurses, and technicians outperform AI when it comes to identifying symptoms and assigning treatments. But venture capitalist Robert Mittendorff, whose firm has invested in companies that make hospital software and online therapies, predicts that machines will eventually outpace niche specialists.
Will a derm AI mean fewer jobs for human skin specialists? Not just yet. More than 5 million Americans require treatment for skin cancer each year, and there are fewer than 12,000 practicing dermatologists in the U.S.