Screening for skin cancer is an imperfect science, but an international team of scientists believes AI can help boost the test's accuracy. As they report in a study published in the Annals of Oncology, a machine-learning program known as a deep learning convolutional neural network (CNN) can be trained to recognize skin cancer with a greater success rate than professional dermatologists.
Researchers taught the CNN to identify skin cancer by showing it more than 100,000 images of malignant melanomas and benign moles. "The CNN works like the brain of a child," co-author Holger Haenssle, senior managing physician at the University of Heidelberg, said in a statement. That means the more information it's given about a certain task, the more it can learn and fine-tune its performance.
After training the AI with a database of images, the researchers showed it a different set of images it had never seen before. The CNN correctly diagnosed skin cancer from images alone 95 percent of the time. When 58 dermatologists were given the same task, the were able to catch only 86.6 percent of the malignant melanomas. The CNN was also less likely to misdiagnose a benign mole as cancerous.
The results don't necessarily mean that AI robots will be replacing flesh-and-blood doctors (or even pigeons) for cancer screenings in the near future. Rather, the researchers see the program acting as a supplement to dermatologists in the clinic, perhaps by evaluating images already stored in the doctors' databases and generating "expert opinions" on the likelihood of cancer.
Even as a doctor's aid, the CNN in its current state leaves room for improvement: The images it looked at were mostly of white patients that didn't include the full range of skin lesions. Diagnosing melanomas that show up on fingers, toes, and scalps also presents a challenge when working with an image-based system. Nonetheless, the researchers are confident that these issues won't stop AI from playing a role in future cancer screenings. "Given exponential development of imaging technology, we envisage that sooner than later, automated diagnosis will change the diagnostic paradigm in dermatology," researchers said.