Your phone might someday save your skin.
Stanford researchers say they've created a new artificial intelligence system that can identify skin cancer as well as trained doctors can. According to a study they published in science journal Nature, the program was able to distinguish between cancerous moles and harmless ones with more than 90 percent accuracy.
The researchers trained the system by feeding it nearly 130,000 images of moles and lesions, with some of them being cancerous. The system scanned the images pixel by pixel, identifying characteristics that helped it make each diagnosis. Using machine learning, the A.I. grew more accurate as it studied more samples.
It then went head to head with 21 trained dermatologists. The result: The A.I. software achieved "performance on par with all tested experts." The system correctly identified 96 percent of the malignant samples, and 90 percent of the (generally harmless) benign ones. For the doctors in the study, those numbers were 95 percent and 76 percent, respectively.
This could have huge implications: The study points out that 5.4 million new cases of skin cancer are diagnosed each year in the U.S. alone. If installed in smartphones, the authors say, this technology could provide a simple, low cost form of early detection.
Identifying melanoma early on is critical. The five-year survival rate when the cancer is caught in its earliest stages is 99 percent. That number drops to 14 percent when detected in its late stages. Having the equivalent of a dermatologist--as far as diagnosing goes--in your pocket could help patients keep a closer watch on their own skin and seek medical treatment sooner.
That's not to say dermatologists will be replaced--they'd still be the ones to perform any procedures necessary. And in a blog post on Stanford's website, the authors suggest doctors might use the tool for in-office diagnoses.
Before the system can achieve its potential, though, it will have to be able to detect cancer from images captured by smartphones. While phone cameras are rapidly improving, the A.I. is currently trained to work only with high quality medical images.
Still, the technology is moving in that direction. Being able to detect early could have an impact on the 10,000 people who die from skin cancer each year in the U.S. alone.
The Stanford researchers developed the framework for the A.I. system using an image classification algorithm that had previously been built by Google.