An artificial intelligence algorithm powered by a database of more than 130,000 images of skin lesions can detect skin cancer as well as a human dermatologist, researchers say.
The AI, developed by a team from Stanford University, has a few major differences from the typical advanced neural networks that have seen increasing use among medical researchers. For one, it's not built from scratch, but instead harnesses an existing Google database of 1.28 million objects. That means it already knows how to classify things, so the researchers could spend less time on training the algorithm and more on fine-tuning its sensitivity and accuracy.
The end result is a computing tool that could drastically simplify the process of diagnosing skin cancer in areas of the world that don't have hospitals or clinics. The algorithm was tested against the diagnoses of 21 board-certified dermatologists, and the researchers said it matched the humans' performance.
Since the first step in skin cancer detection is a visual inspection, theoretically all the AI would need is a good-enough quality photo of the patient, perhaps taken with a smartphone. There are still a few wrinkles to iron out before that happens, though, the most important of which is that the algorithm is currently designed for supercomputers, not smartphones. It also has yet to pass clinical trials. Still, its creators are optimistic.
"Advances in computer-aided classification of benign versus malignant skin lesions could greatly assist dermatologists in improved diagnosis for challenging lesions and provide better management options for patients," Stanford professor of dermatology Susan Swetter said in a statement. "However, rigorous prospective validation of the algorithm is necessary before it can be implemented in clinical practice, by practitioners and patients alike."