The versatility of the smartphone is starting to have a serious impact in the medical world, with the ability to reveal low red blood cell counts, sleep apnea and even HIV all very real possibilities. Skin cancer too is a condition that might feel the wrath of these modern diagnostic tools, with an iPhone app way back in 2011 offering risk assessments on suspect moles. But a new research project at Stanford University is promising to bring things up to a professional grade of diagnosis, through a deep learning algorithm that can detect potential cancers with the same accuracy as dermatologists in early tests.
As is the case with all cancers, early detection of skin cancer is critical to survival rates. For melanoma detected in its early stages, the five-year survival rate is 97 percent, but those detected in its later stages carry a survival rate of just 14 percent. However, not everybody has ready access or the funds to drop by the doctor's office and get their skin oddities checked out as soon as they appear.
Looking to improve the odds in these cases, computer scientists from Stanford University set out to build an artificially intelligent algorithm that uses deep learning to detect early-stage skin cancers. In general terms, deep learning means that instead of developing a system already loaded with the answers, the system is given the ability to work out the problem on its own.
In this case, this lead the researchers to build a database of almost 130,000 images of skin lesions representing more than 2,000 different diseases. They then took a Google-developed algorithm designed to distinguish cats from dogs, and adapted it to their skin cancer problem by feeding it each image as raw pixels and an accompanying disease label.
"There's no huge dataset of skin cancer that we can just train our algorithms on, so we had to make our own," said Brett Kuprel, co-lead author of the paper. "We gathered images from the internet and worked with the medical school to create a nice taxonomy out of data that was very messy – the labels alone were in several languages, including German, Arabic and Latin."
The team then had 21 trained dermatologists diagnose cancerous and non-cancerous lesions from over 370 images. It then put its new algorithm to the test, tasking it with identifying the most common skin cancers, and then separately identifying the deadliest of skin cancers: malignant melanomas. It found that in both tasks, the algorithm's performance was on par with the experts, demonstrating that the AI could classify skin cancers on a comparable level to trained dermatologists.
"We realized it was feasible, not just to do something well, but as well as a human dermatologist," said Sebastian Thrun, an adjunct professor in the Stanford Artificial Intelligence Laboratory. "That's when our thinking changed. That's when we said: 'Look, this is not just a class project for students, this is an opportunity to do something great for humanity.'"
For the researchers, getting this kind of technology into as many hands as possible involves only one avenue: the modern smartphone with its ever-improving camera and various sensors. While the algorithm was developed and currently exists on a computer, the team believes it could be adapted for the smartphone without too much trouble, and what a breakthrough that would be. There is still some serious testing to be done before that can happen, but imagine if anytime a suspicious skin marking had you feeling uneasy you could whip out your smartphone and get some answers. Doctors at dinner parties rejoice!
"Advances in computer-aided classification of benign versus malignant skin lesions could greatly assist dermatologists in improved diagnosis for challenging lesions and provide better management options for patients," said Susan Swetter, professor of dermatology at the Stanford Cancer Institute, and co-author of the paper. "However, rigorous prospective validation of the algorithm is necessary before it can be implemented in clinical practice, by practitioners and patients alike."
The research was published in the journal Nature.
Source: Stanford University
