Can AI Diagnose Melanoma Better than a Doctor?
What if a machine could “see” what doctors can’t? Soon, there may be a an artificial intelligence tool for diagnosing melanoma that can do just that. Researchers at the University of California and Stanford University are developing a scanner and an app for your smart phone that may provide earlier melanoma detection than a melanoma biopsy.
AI Coming to Skin Cancer Detection
Melanoma represents only 5% of skin cancers. But it causes the most skin cancer deaths. Why? It’s not easy to spot and treat it in its earliest stages—when melanoma is nearly 100% curable. Early detection is the key to survival.
The current standard of care is visual inspection by a clinician. But even if you do monthly skin exams, you may not notice a change. And, if a patient has hundreds of moles, it can be difficult to catch and treat before it becomes a serious health issue.
A Computer to Rival the Human Brain
Inspired by how the human brain works, these scientists are taking advantage of major gains in computer power and big data. By comparing a picture of a mole or lesion to images gathered from the world wide web, neural networks “models” are trained to predict whether a mole is melanoma based.
At Stanford, computer scientists “trained” an existing Google algorithm to combine visual processing with a type of artificial intelligence called deep learning. The images used in their app represent over 2,000 different skin diseases, gathered from the internet and vetted by dermatologists.
At UC, The algorithm is programmed to look across thousands of images of skin lesions and observe patterns. This data is then combined with information collected from a database of melanoma patients to create a scanning tool. Both Stanford and UC researchers involved with these projects believe that this new technology will be able to “read” your skin better than a clinician.
UC Santa Barbara undergrad Abhisheck Bhattacharya, who developed the melanoma project along with UC San Francisco professor, Dr. Dexter Hadley, has so far proven a 96% accuracy rate using this model.
“We’re applying computer vision to solving medical problems,” said Bhattacharya.
“What we’re trying to do is something humans can’t do,” says Dr. Hadley. “Humans can’t predict what goes on under the skin from looking from the top.”
Regardless of the accuracy rate in the laboratory, researchers at UC are taking steps toward proving the success rate in a clinical trial before the technology is introduced to the public.
At Stanford, researchers found that their app is also capable of identifying skin cancer as well as dermatologists (91% accuracy). The Stanford team says the aim of developing their program is not to replace human dermatologists, but to offer people an inexpensive option for early screening. Both teams say that the model neural network looks promising, however, more rigorous assessments of its safety would need to be made before such a program could go public.
Have a questions about your skin? Consult a dermatologist now.
A Pain-Free Alternative to Biopsy
Traditionally, a full thickness biopsy is the only way to know for sure how deeply a malignant melanoma has penetrated the skin. Most of the time, the physician may find it necessary to perform multiple biopsies. With more accurate melanoma detection, patients may benefit by having to endure fewer biopsies.
UC and Stanford teams are focused on bringing artificial intelligence (AI) to skin cancer detection soon. Tools that allow anyone, especially those with little access to skin experts, to assess whether they may have signs of skin cancer will likely be life-saving. The earlier melanoma is detected, the better the chances to beat it.
For now, the fastest way to get a mole checked out is using First Derm’s service. Bypass the average wait time of 32 days for a dermatologist appointment and get an answer within hours.
Ask a Dermatologist
Anonymous, fast and secure!
Specialist doctor from the University Hospital in Gothenburg, alumnus UC Berkeley. My doctoral dissertation is about Digital Health and I have published 5 scientific articles in teledermatology and artificial intelligence.