Researchers have demonstrated a computer-driven image recognition system capable of diagnosing many types of skin cancer as accurately, they claim, as dermatologists, according to a paper published online in Nature (Jan. 25, 2017).
Developed at the Stanford Artificial Intelligence Laboratory at Stanford University in California, the computer system—known as a convolutional neural network (CNN)—was created to see if it was practical to use an automated system to take over some of the diagnostic work from physicians. The team built a database of nearly 130,000 skin disease images and trained their algorithm to visually diagnose potential cancer. In initial tests, it demonstrated high accuracy. “We realized it was feasible, not just to do something well, but as well as a human dermatologist,” said Sebastian Thrun, senior author on the paper and an adjunct professor in the laboratory, in a press release.
A total of 129,450 digital images and their corresponding disease, which encompassed 2,032 different diseases, were input into the computer system.
“There’s no huge dataset of skin cancer that we can just train our algorithms on, so we had to make our own,” Brett Kuprel, co-lead author of the paper and a graduate student in the Thrun lab, said in the release. “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.”
Then the algorithm’s diagnostic accuracy was tested against 21 board-certified dermatologists on high-quality, biopsy-confirmed images provided by the University of Edinburgh and the International Skin Imaging Collaboration Project. Two binary classification use cases were examined: keratinocyte carcinomas versus benign seborrheic keratoses; and malignant melanomas versus benign nevi. Those two cases, the authors note, represent the identification of the most common cancers and the identification of the deadliest skin cancer, respectively. Diagnostic accuracy was assessed through three tasks: keratinocyte carcinoma classification, melanoma classification, and melanoma classification when viewed using dermoscopy. In all three tasks, the algorithm matched the performance of the dermatologists.
While the team that developed the algorithm believes it will be relatively easy to transition the algorithm to mobile devices, they note that further testing in a real-world clinical setting is needed.
“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 Dr. Susan Swetter, professor of dermatology and director of the pigmented lesion and melanoma program 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.”