When you hear the term “image recognition technologies,” you might think of CIA surveillance or James Bond, not high-tech medicine. But a team of researchers from Stanford have used artificial intelligence healthcare technology to identify images of skin cancer moles and lesions as accurately as a dermatologist in many cases. The new study was published in the journal Nature.
While medical AI technology is still in its infancy, it shows a lot of promise. The new study suggests that in the future, a simple smartphone application could help patients diagnose skin cancer (the most common cancer in the U.S.) by themselves.
Catching Skin Cancer Early
While melanomas represent fewer than 5 percent of skin malignancies diagnosed in the United States, they account for almost 75 percent of deaths related to skin cancer. With early detection, the five-year survival rate is 99 percent. If skin cancer isn’t detected until Stage IV, the survival rate is only 14 percent. In the past, dermatologists had to identify whether moles or other skin abnormalities were cancerous by looking at them.
Sebastian Thrun, senior author of this new study, an adjunct at Stanford and founder of Google X, developed a deep learning computer system to identify skin cancer at a glance. By creating this AI cancer diagnosis tool, Thrun and his team essentially developed a computer-based dermatologist.
How Does this System Work?
The researchers in the study started by coaching the computer to develop pattern recognition skills—essentially using algorithms for deep learning through a convolutional neural network. By teaching the algorithm what the world looked like, they could train it to understand different skin conditions, even though they look different on individual people.
In total, the Stanford researchers fed the program almost 130,000 images representing more than 2000 skin diseases. This wealth of information allowed the program to be skilled at identifying the first diagnosis of skin diseases.
 |
| Early detection of skin cancer is key to effective treatment. Source: cancer.gov |
How Does This Differ from a Real Doctor?
The major difference between artificial intelligence healthcare and a human doctor is that humans can learn to recognize skin cancer patterns from just several examples, while the computer needs thousands or even billions of examples. Ultimately, though, the computer algorithm was just as successful as board-certified dermatologists at several key diagnostic tasks.
What Does This Mean for Medicine?
Despite the program’s skill, however, real testing in a clinical setting is still necessary after initial diagnosis. The study’s researchers believe that their research could be extended to other fields, including ophthalmology, radiology, and pathology. If a smartphone app can achieve a primary diagnosis, introducing artificial intelligence in medical diagnosis brings us one step closer to low-cost universal access to diagnostic care.
As smartphone use throughout the world increases, tools like this could be useful in diagnosing patients in areas without access to dermatologists and then bringing those patients in for care if necessary. Artificial intelligence medicine combines technology and human knowledge in a way could help bridge the health care gap that exists in low-income neighborhoods in the United States and developing countries throughout the world.
