Researchers at Stanford University trained a deep neural network on 129,450 clinical images spanning 2,032 different skin diseases, then tested it against 21 board-certified dermatologists reading the same photographs.
The network matched the dermatologists' accuracy at telling malignant lesions from benign ones, across both the most common skin cancers and the deadliest, melanoma, using only an image with no biopsy, no history, and no exam of the patient in front of it.
The model only replicated what 21 dermatologists saw in a photograph; it has not replaced a biopsy, which remains how a diagnosis is confirmed, and image quality, lighting, and skin tone all affect what any image-based system can read. The dataset skewed toward lighter skin, a gap later studies pushed to correct. The result still pointed toward a camera doing a pass that used to need a referral.
Read the original study
Twenty-one dermatologists, matched by a model.
Source
- Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H. M., & Thrun, S. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115–118. https://doi.org/10.1038/nature21056