The system was developed to help doctors determine the stage that a skin cancer has reached. While patients often independently find melanomas by spotting a new mole or a change in an existing one, even dermatologists can struggle to decide whether it’s invasive or not. The researchers suspected that AI could assist them with the task. [Read: How Polestar is using blockchain to increase transparency] They classified the melanomas with a convolutional neural network (CNN), a powerful method of analyzing images that’s proven adept at identifying different skin lesions. The researchers trained and validated the CNN on 937 images of melanomas collected through a dermascope, a handheld instrument used to examine the skin. They then tested the algorithm’s evaluations on 200 cases that had been diagnosed by a dermatopathologist. When they compared its performance to the analysis of seven independent dermatologists, the result was a draw. “None of the dermatologists significantly outperformed the ML algorithm,” said study author Sam Polesie. The researchers acknowledge that the algorithm still needs further refinement and longer-term evaluation in a clinical setting. However, their study shows that AI could help assess the severity of melanoma before surgery, which affects how extensive an operation needs to be. You can read a pre-proof of the study paper in the Journal of the American Academy of Dermatology.
