A new method of detection relies on artificial intelligence
That may sound scary, but the gist of it is pretty simple. After showing a computer program a massive library of clinical images of various skin cancer moles and lesions, Stanford scientists have taught a machine to correctly identify skin cancer as accurately as a dermatologist would.
"Our objective is to bring the expertise of top-level dermatologists to places where the dermatologist is not available," Sebastian Thrun, the senior author of the study, told CNN. (Thrun is also the founder of Google X, a semi-secret research and development facility responsible for other AI projects, as well as Google's self-driving car.)
They did it by applying a "deep convolutional neural network" -- basically a packet of algorithms used in artificial intelligence -- to a dermatologist's regular diagnostic process. The first thing a trained dermatologist will do is look at a patient's moles or skin lesions and tell them whether the mole should be biopsied for skin cancer. In essence, these scientists "taught" their convolutional neural network how to do the same.
After priming the algorithm over the course of a week with millions of images of ordinary items -- dogs, tables, chairs, etc. -- they then fed it a dataset of 129,450 clinical images of verified, biopsy-proven skin aberrations. These represented more than 2,000 different types of skin diseases and aberrations. All the while, the neural network took in that information and wrote -- "learned" -- rules around how to diagnose those skin aberrations.
Then came a diagnostic test on both the deadlier melanoma and non-melanoma skin cancers. The researchers put their convolutional neural network up against 21 board-certified dermatologists and found its diagnoses to be just as accurate as theirs. Per the study: "The CNN achieves performance on par with all tested experts across both tasks, demonstrating an artificial intelligence capable of classifying skin cancer with a level of competence comparable to dermatologists."
Which is kind of insane. The study points out that "this fast, scalable method is deployable on mobile devices." It could change the game of how skin cancer, and melanoma specifically, is treated. If you catch melanoma early, your five-year survival rate is 99%. If you catch it in its latest stage that number drops to 14%. We're not quite there yet, but imagine being able to diagnose it on your phone.
"This is a very specific study, and it has a very encouraging result," said Thrun, but he admits that they'll need to test it for real in doctors offices before it can start the real work of saving lives.