Skin cancer detection won’t be turned over to machines anytime soon, but artificial intelligence detected skin cancer more accurately than a large group of international dermatologists in controlled testing, Agence France Presse reports.
In an academic study and clinical trial published in Annals of Oncology, the study’s lead author, Professor Holger A. Haenssle, of the University of Heidelberg Department of Dermatology, wrote, “Most dermatologists were outperformed by the CNN. Regardless of any physician’s level of experience, they may benefit from assistance by a CNN’s image classification.”
Man versus machine
Prior to the test, researchers from Germany, France, and the U.S. taught the CNN to differentiate benign skin lesions from dangerous melanomas. In the process, the team showed more than 100,000 images of correctly identified skin cancers to the neural network, which was designed with Google’s Inception v4 CNN architecture.
The 58 dermatologists were divided into three self-identified groups: beginners with less than two years of experience, skilled with two to five years, and experts with more than five years of experience. There were 19 beginners, 11 skilled, and 30 experts among the group.
Two tests were run. In one test the dermatologists were shown 100 dermoscopic images with no other information. They were asked to indicate whether the cancer was a melanoma or benign. In addition, the doctors were asked whether they would recommend excision, short-term follow-up, or no action. Four weeks later the dermatologists were shown the same images again, this time with additional clinical information about the patients plus close-up images.
The CNN scored higher than the overall group of dermatologists on both tests, with and without extra information. The dermatologists accurately identified an average of 86.5 percent of the skin cancers on the image-only test. In the second test, with more information, the doctors averaged 88.9 percent accuracy. The CNN, however, correctly detected the types of cancers 95 percent of the time based on images only.
Rated by experience group, none of the three groups of dermatologists was as accurate as the neural network. The team did report, however, that 18 of the dermatologists scored higher than the CNN.
“The CNN missed fewer melanomas, meaning it had a higher sensitivity than the dermatologists,” Haenssle said. It also “misdiagnosed fewer benign moles as malignant melanoma … this would result in less unnecessary surgery.”
According to the authors of the study, the test does not mean machines will replace doctors. One issue is that melanomas can be difficult to recognize or image in some parts of the body such as the toes and scalp. The study calls for repeated, large-sized clinical tests.
The test does show, however, that dermatologists at all skill levels could benefit from A.I. assistance in skin cancer classification.
- Can A.I. beat human engineers at designing microchips? Google thinks so
- New A.I. can identify the song you’re listening to by reading your brain waves
- Wild new ‘brainsourcing’ technique trains A.I. directly with human brainwaves
- Deep-learning A.I. is helping archaeologists translate ancient tablets
- According to advanced swarm A.I., this is who will win Super Bowl LIV