Skin disease artificial intelligence (AI), which judges the skin condition from skin images and support medical treatments, are highly expected to be a bridging device that complements skin condition judgment by the eyes of medical professionals. Image AI development requires many high-quality images with correct diagnostic information. It is essential for skin disease AI development to use a skin image database that has been precisely diagnosed and accumulated by dermatologists. Japanese Dermatological Association (JDA) has started to construct the National Skin Disease Database (NSDD) by support of the AMED grant since 2018. In March 2021, NSDD composed of a half million skin disease images and skin pathology images selected by the dermatologists. NSDD is still expanding to collect a million skin disease images.
A part of NSDD has been used to develop a skin disease AI by the collaboration with National Institute of Informatics. From the accumulated skin images, we first selected 59 diseases with a high frequency and a large number of images. The 59 diseases include variety of skin diseases, such as skin tumors, inflammatory skin diseases, and skin infection. The developed AI classifies the 59 skin diseases well: 0.582 accuracy in Top 1, 0.793 accuracy in Top 3, and 0.863 accuracy in Top 5 diagnosis. The results proved that NSDD is useful to develop accurate skin disease AI.