In the not-so-distant future, you may be able to skip the dermatologist visit and instead use a smartphone app to check if you have skin cancer.
Researchers at Stanford University trained a computer to identify images of cancerous moles and lesions as accurately as a dermatologist, according to a study published in the journal Nature.
“Our objective is to bring the expertise of top-level dermatologists to places where the dermatologist is not available,” said Sebastian Thrun, senior author of the new study, founder of research and development lab Google X, and adjunct professor at Stanford University.
Many people in developing countries do not have access to good medical care, and this new technology could potentially be life-saving. Experts believe by 2021, there will be roughly 6.3 billion smartphone subscriptions. The researchers hope that their new system in the form of an app could provide low-cost access to diagnostic care.
Less than one percent of all skin cancer cases in the U.S. are diagnosed as melanoma, but it counts for about three-quarters of all deaths related to skin cancer. If caught early, the five-year survival rate is 99 percent. However, if detected late, the survival rate drops to just 14 percent.
Dermatologists are able to identify when a mole or other skin abnormality is cancerous just by looking at it. Typically, they perform a follow-up biopsy and tests to confirm their diagnosis.
Thrun and his team developed a deep learning computer system — or an algorithm-based technique — that could identify skin cancer at just a glance.
“An algorithm is just a fancy name for a sequence of steps that the computer takes. So in this case, the algorithm refers to the whole process that they did to train the system,” Carl Vondrick, a Ph.D. candidate at MIT’s Computer Science and Artificial Intelligence Lab, who was not involved in this study, told CNN.
“We taught it with [sp] cats and dogs and tables and chairs and all sorts of normal everyday objects look like,” Andre Esteva, co-first author of the new paper and Electrical Engineering Ph.D. student at Stanford, said. “We used a massive data set of well over a million images.”
Once this phase of learning was complete, Esteva trained the algorithm to recognize different skin conditions.
At this point in the process, the team of researchers were confronted with another complicated problem — cancerous and noncancerous skin aberrations vary in appearance from patient to patient. In order to overcome this hurdle, the researchers uploaded an extensive dataset of 129,450 images containing more than 2,000 skin diseases. The computer was also fed the diagnoses of each image of a mole or skin abrasion.
“[The computer was able to] diagnose multiple different kinds of skin cancer, not just melanoma, and we were able to do this with regular clinical images, rather than with specialized dermoscopic images,” said Roberto Novoa, a co-author of the study and dermatologist at Stanford Medicine.
The researchers agree that the real testing will need to be in a clinical setting, but believe their work might expand to other areas of medicine such as ophthalmology, radiology and pathology.