Has that mole always been so dark? Has that pink bump been there too long? Did I always have a freckle there? If you’ve ever asked yourself questions like these, then you understand how frightening the thought of skin cancer can be.
If you’re like most of the population, you’ve been putting off that dermatology appointment for months. But what if you could get that mole checked out without even leaving your bedroom? It may soon be possible.
A team of Stanford University researchers has developed a computer algorithm that can detect skin cancer with the same level of accuracy as leading dermatologists. Once the stuff of science fiction, this remarkable research suggests that we are on the cusp of a new age of medicine, one populated by artificially intelligent dermatologists.
Early Detection Saves Lives
Skin cancer is the most common form of cancer in the United States — approximately one in five Americans will be diagnosed with it at some time in their lives. Non-melanoma skin cancer, the most common type of skin cancer, is almost completely curable when it is detected early and treated properly. Melanoma, though relatively rare, is also curable if caught early, but does cause most skin cancer fatalities. It is a far more aggressive type of cancer, which means that early detection can make the difference between life and death.
According to the American Cancer Institute, melanoma in its earliest stage has a five-year survival rate of 97 percent. However, that number drops precipitously as the cancer progresses. In its latest stage, melanoma has a five-year survival rate of less than 20 percent. Most of these deaths can be prevented by early detection, but too many people around the world (and here in the U.S.) lack access to preventative care services, like skin cancer screening.
A Different Approach
The medical community has been teaching us for years about ways to prevent skin cancer, and for the most part, people have followed their advice. We have forgone tanning beds. We cover up. We use SPF every day. But still, we forget to get annual skin checks. So, how do we make skin cancer screening simpler and more accessible?
Sebastian Thrun, a computer scientist and the former director of the Artificial Intelligence Lab at Stanford University, set out to answer this very question. Knee deep in self-learning robots and driverless cars, Thrun began to reflect on the death of his mother, who was taken by breast cancer when he was young.
“I became obsessed with the idea of detecting cancer in its earliest stage—at a time when you could still cut it out with a knife,” Thrun told The New Yorker. “And I kept thinking, ‘Could a machine-learning algorithm help?’”
Deep learning — essentially, a computer that can learn — is promised to be the future of technology. Today, an MRI machine that has seen 2,000 tumors understands no more about tumors than it did on the day it was first powered on. The computer system Thrun and his team of researchers developed, however, is teachable. Each time it sees a picture of a mole, it applies that knowledge to its general understanding of benign and malignant skin conditions.
The Artificial Intelligence Is In
Thrun’s new research, published in Nature, documents the process by which he and his team taught their machine to diagnose skin cancer. They began by teaching it basic child-like things, such as how to recognize pictures of dogs and cats. Then they began focusing on the most common types of skin cancer (non-melanoma) and the deadliest (melanoma).
The objective of this new technology is to one day diagnose skin cancer through a smartphone app. To meet that objective, the machine learned to identify pictures with a variety of different focus levels, angles and lighting conditions.
After a short time in “school,” Thrun’s dermatological AI was pitted against a group of 21 board-certified dermatologists. Both the AI and the dermatologists were given roughly 2,000 images of biopsy-proven skin conditions and asked to provide a diagnosis. The results were staggering.
The machine was just as accurate as the dermatologists on most tests, even outperforming them in some levels of specificity. For instance, the computer was better at not calling something melanoma when it wasn’t, which is a common problem that leads to unnecessary biopsies.
The applications for this deep learning technology are huge. Thrun’s hope is that within the next few years people will be able to snap a picture of a worrisome mole and get a diagnosis from their smartphone. This technology won’t be replacing dermatologists anytime soon, but it definitely has the potential to streamline the process.
That Sounds Great, But What Can I Do Today?
Artificial intelligence may be coming to a smartphone near you, but it is not here yet. Our devices are not yet capable of making any kind of skin care diagnosis. There are a few apps out there that falsely claim to be able to diagnose skin cancer, but they are not to be trusted. Some claim to use various types of image analysis to determine your level of risk, but they are not backed up by research. No available technology can diagnose skin cancer. If you are worried about a spot, bump or mole on your body, go to the dermatologist.
There are, however, some apps already out there that can help you prevent skin cancer by managing your sun exposure, tracking your moles, and reminding you to get skin screenings.
The Mole Mapper: Developed by the Oregon Health and Science University, this iPhone app lets users take photos of their skin and monitor their moles over time. The Mole Mapper captures the size, shape and color of moles so that people can more easily notice unusual changes. This app is actually part of a research study that collects data and images of skin conditions and makes them available to researchers like Sebastian Thrun.
MySkinPal: This is a mole tracking and analysis app that helps you keep track of changes over time. Users can take photos of moles every few weeks then play back the images to look for changes. You can also apply color filters to your images. The app sends you a reminder when you haven’t scanned a mole in a while. Your data is kept confidential, but can easily be shared with your dermatologist if you choose. You can keep track of family member’s moles too!
My UV Patch: La Roche-Posay has introduced a flexible, wearable technology that looks like a little Band-Aid. It works in conjunction with your smartphone to determine your level of UV exposure. The patch has photosensitive dyes that change color when exposed to certain amounts of UV light. You use your phone to scan the patch and receive personalized advice about your sun exposure, such as when to re-apply SPF.