Researchers have developed artificial intelligence that could be just as good at detecting skin cancers as dermatologists.
The team of computer scientists from Stanford University set out to produce an algorithm for diagnosing skin cancer, making a database of almost 130,000 skin disease images and training the algorithm to visually diagnose the condition.
The finding: the AI very well identified malignant carcinomas and melanomas as much as 21 dermatologists can.
“We made a very powerful machine learning algorithm that learns from data,” said Stanford graduate student and study author Andre Esteva in a statement, explaining that the algorithm figures out what to look for in order to visually identify skin cancer.
Identifying Melanomas and Carcinomas
The algorithm — developed by Google and already trained to figure out 1.28 million images from a thousand object categories — was fed each of the 129,000 or so images as raw pixels with a corresponding disease label, which could be in languages like German and Arabic. The mission was for the algorithm to differentiate a malignant carcinoma from a benign skin lesion.
The 21 dermatologists were tasked to diagnose cancerous and non-cancerous lesions in more than 370 images. In all three tasks, namely keratinocyte carcinoma classification, melanoma classification, and melanoma classification, the algorithm successfully matched the doctors’ performance.
The team seeks to make the algorithm smartphone-compatible in the near future in order for reliable yet affordable skin cancer diagnosis. While they believe it will be quite easy to shift the algorithm to a mobile setting, they would still need to conduct added testing in real-world clinical environments.
The visual diagnosis witnessed in the study, detailed in the journal Nature, is hoped to be replicated in other medical fields soon.
The Scourge Of Skin Cancer
In the United States alone, there are around 5.4 million new cases of skin cancer every year. The five-year melanoma survival rate, when the condition is found in its earliest stage, is about 97 percent, but this falls to about 14 percent when detected in the late stages.
Skin cancer diagnosis typically starts with a visual exam, where a dermatologist looks at the lesion in question under a dermatoscope. If the technique is proven inconclusive or if the doctor deems the lesion cancerous, a biopsy will be conducted next.
In Australia, physicians have more than 1 million patient consultations revolving around skin cancer each year, and they comprise almost 80 percent of all newly diagnosed cancers.
But while saying AI research could be “the way of the future,” Aussie dermatologist Dr. Natasha Cook warned that machines can still make errors.
"You can't be completely reliant on technology, because there are certain things that will actually defy standards and require a history and require human input, so that you don't miss things," explained Cook.
Excessive sun exposure has been implicated in skin cancer, where mounting research warns against getting exposed to harmful ultraviolet rays from the sun and tanning beds. In a recent study, sunscreen provided stronger sun protection than beach umbrellas when it comes to extended UV exposure.