AI outperforms doctors at skin cancer diagnosis
Image credit: Dreamstime
An international team of researchers have demonstrated the superhuman ability of an artificial intelligence (AI) to detect the most lethal form of skin cancer – melanoma – for the first time.
Malignant melanoma is on the rise, with 232,000 new cases and 55,500 deaths estimated each year. While the disease can be treated with surgery in its earlier stages, often diagnoses are not made until the disease has advanced.
In order to see whether AI could play a role in crucial early detection, the French, German and American researchers trained a convolutional neural network (CNN) – an algorithm inspired by the structures and processes of a brain often used for processing images – with a huge database of images of benign moles and malignant melanoma.
“The CNN works like the brain of a child. To train it, we showed the CNN more than 100,000 images of malignant and benign skin cancers and moles and indicated the diagnosis for each image” said Professor Holger Haenssle, a senior doctor at the University of Heidelberg’s Department of Dermatology.
“With each training image, the CNN improved its ability to differentiate between benign and malignant lesions.”
Next, the researchers generated two new sets of images to test the CNN, with some of the images being sent to 58 dermatologists from 17 countries for their analysis. The dermatologists made diagnoses and – if appropriate – suggested a course of treatment. After a few weeks, they were given more information about the patient – as well as the locations of the regions depicted on their bodies – and asked for their opinions once again.
In the first round, the dermatologists were able to detect 86.6 per cent of instances of skin cancer, and 71.3 per cent of benign moles. When tuned to detect the same percentage of benign moles as the dermatologists, the CNN was able to detect 95 per cent of melanomas. After the second round, the dermatologists improved their correct positive diagnoses slightly to 88.9 per cent.
“The CNN missed fewer melanomas, meaning it had a higher sensitivity than the dermatologists, and it misdiagnosed fewer benign moles as malignant melanoma […] this would result in less unnecessary surgery,” said Haenssle.
“These findings show that deep learning convolutional neural networks are capable of outperforming dermatologists, including extensively trained experts, in the task of detecting melanomas.”
While Haenssle and his colleagues do not suggest that a similar CNN should take over the job of dermatologists in detecting skin cancer, they suggest that it could be a useful tool in helping with early detection.
Earlier this month, Prime Minister Theresa May laid out government plans for integrating AI into cancer care, with the intention of using these data-based tools to identify the disease at the earliest possible stage to improve patients’ chances of survival.