Artificial intelligence might be better than dermatologists at correctly identifying skin cancer, according to new research. The findings could help improve strategies to detect cancerous skin lesions early on and provide faster, more effective treatment, researchers say.
Researchers in Germany, the United States and France used a machine learning system called a deep learning convolutional neural network (CNN) for the study. The team showed the CNN over 100,000 magnified images of benign moles, known as nevi, and malignant melanomas, the most deadly form of skin cancer, in order to train the system how to tell the difference.
The study, published on Monday in the journal Annals of Oncology, involved 58 dermatologists from 17 different countries. Over half of the participating dermatologists had more than five years of experience, 19% of the doctors had between two and five years of experience and 29% had less than two years of experience.
Researchers then showed a set of 300 new images to the CNN and 100 of the most difficult images from that set to the group of dermatologists. When asked to make a diagnosis of melanoma or benign mole solely from the images, the CNN detected 95% of melanomas compared to around 87% correctly identified by dermatologists. Both the CNN and dermatologists correctly identified around 71% of benign moles.
“The CNN missed fewer melanomas, meaning it had a higher sensitivity than the dermatologists, and it misdiagnosed fewer benign moles as malignant melanoma, which means it had a higher specificity,” study author Holger Haenssle, senior managing physician at the Department of Dermatology at the University of Heidelberg, Germany, said in a statement. “This would result in less unnecessary surgery.”
The dermatologists’ scores improved slightly in a second task – when they were given additional clinical information about the patients, doctors correctly diagnosed malignant melanomas in about 89% of cases.
The researchers said that they do not think the CNN would replace the role dermatologists play in diagnosing skin cancers, but rather that the technology could provide additional support.
“This CNN may serve physicians involved in skin cancer screening as an aid in their decision whether to biopsy a lesion or not,” said Haenssle. “Most dermatologists already use digital dermoscopy systems to image and store lesions for documentation and follow-up. The CNN can then easily and rapidly evaluate the stored image for an ‘expert opinion’ on the probability of melanoma.”
Cases of malignant melanoma as well as non-melanoma skin cancers are becoming more common around the world. Each year, there are an estimated 232,000 new cases of malignant melanoma worldwide and 55,500 deaths caused by the disease. Although it can be cured if detected early, many cases are not diagnosed until later, when the cancer is more advanced and therefore more difficult to treat.
Researchers involved in the new study say their findings could improve early detection strategies.
“I have been involved in research projects that aim at improving the early detection of melanoma in its curable stages for almost 20 years,” explained Haenssle. “My group and I are focusing on non-invasive technologies that may help physicians not to miss melanomas, for instance, while performing skin cancer screenings.”
Haenssle added that the team is “currently planning prospective studies to assess the real-life impact of the CNN for physicians and patients.”