Dermatologist-level classification of skin cancer with deep neural networks

What the paper explores

Esteva and colleagues trained a deep neural network on a large collection of clinical skin images and compared it with dermatologists on selected biopsy-proven classification tasks. The work demonstrated the potential of image-based learning for narrowly defined diagnostic distinctions.

How to interpret the evidence

Performance on curated binary classification tasks is not equivalent to safe, prospective clinical use across all skin diseases or patient groups. A correction exists and is linked below; use the corrected publication when interpreting details.

Questions for MedTech teams

Separate early technical feasibility from clinical validation. Define the target population, image acquisition conditions, workflow and errors that must be evaluated before clinical claims are considered.