Ofer Reiter1,2 & Veronica Rotemberg1 & Kivanc Kose1 & Allan C. Halpern1AbstractPurpose To review recent developments in artificial intelligence for skin cancer diagnosis.Recent Findings Major breakthroughs in recent years are likely related to advancements in utilization of convolutional neural networks (CNNs) for dermatologic image analysis, especially dermoscopy. Recent studies have shown that CNN-based approaches perform as well as or even better than human raters in diagnosing close-up and dermoscopic images of skin lesions in a simulated static environment. Several limitations for the development of AI include the need for large data pipelines and ground truth diagnoses, lack of metadata, and lack of rigorous widely accepted standards.Summary Despite recent breakthroughs, adoption of AI in clinical settings for dermatology is in early stages. Close collaboration between researchers and clinicians may provide the opportunity to investigate implementation of AI in clinical settings to provide real benefit for both clinicians and patients.
