A study in Nature investigates whether a deep learning algorithm can classify skin cancer at comparable rates to trained dermatologists.
The researchers — led by Andre Esteva and Brett Kuprel, both PhD candidates at Stanford (Calif.) University — began with an algorithm developed by Google, which was trained to identify images based on object categories. They went on to train the algorithm to visually identify cancer using a dataset of 129,450 images of 2,032 different skin diseases.
For the study, the researchers tested the performance of the deep learning algorithm against the performance of 21 board-certified dermatologists, in which both groups were asked to classify cancerous and non-cancerous lesions. The deep learning algorithm achieved performance "on par with all tested experts," according to the researchers.
Looking forward, the researchers suggest, "Outfitted with deep neural networks, mobile devices can potentially extend the reach of dermatologists outside of the clinic."