Abstract: 【Abstract】 Conventional machine learning techniques can not be directly used to process natural data in their raw form, and have to rely on experts to design feature extractors. However, the emergence of deep learning has broken this limitation. It is a method that allows a machine to be fed with raw data and to automatically discover representative information needed for detection or classification, and has become a key technology for medical image classification with artificial intelligence. Deep learning has achieved a level comparable to or even higher than that of dermatologists in terms of classification between malignant melanoma and pigmented nevus, as well as classification between skin diseases other than melanocyte-derived tumors, such as squamous cell tumor, basal cell carcinoma and nail disorders. The review introduces some basic concepts of deep learning in skin image classification and common evaluation methods for deep learning models, and summarizes research progress in the application of deep learning in skin image classification.

Key words: Artificial intelligence, Nevi and melanomas, Skin diseases, Dermoscopy, Neural networks （computer）, Skin imaging, Deep learning