Computer 'Deep Learning' Algorithm Matches Dermatologists at Classifying Melanomas
In melanoma classification tasks, a deep learning algorithm performed similar to dermatologists.
A computer algorithm is as accurate as board-certified dermatologists at recognizing and classifying skin cancers using digital image data, according to researchers at Stanford University Artificial Intelligence Laboratory in Stanford, California.1
Using data from nearly 130,000 medical images of skin diseases, the study authors developed a neural-network artificial-intelligence deep learning algorithm that can recognize subtle visual cues of malignancy. They compared the algorithm's performance and that of 21 board-certified dermatologists for 3 diagnostic tasks: melanoma classification, carcinoma classification, and dermoscopic melanoma classification.
The algorithm matched the dermatologists' performance in all 3 tasks.
“Outfitted with deep neural networks, mobile devices can potentially extend the reach of dermatologists outside the clinic,” noted lead study author Andre Esteva, PhD, and colleagues, in a paper published in Nature. “It is projected that 6.3 billion smartphone subscriptions will exist by the year 2021 and can therefore potentially provide low-cost universal access to vital diagnostic care.”
Dermatologists' clinical interpretations include factors “beyond visual and dermoscopic inspection” of skin lesions, the authors were quick to acknowledge. But the algorithm's ability to classify lesion images “with the accuracy of a board-certified dermatologist has the potential to profoundly expand access to vital medical care,” they concluded.
Similar computer deep-learning visual-recognition algorithms have been shown in other settings to outperform humans, the authors noted.
1. Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017 Jan 25. doi: 10.1038/nature21056 [Epub ahead of print]