JMIR Publications

ABSTRACT

Background:

Skin and subcutaneous disease is the fourth leading cause of nonfatal disease burden globally and also one of the most common chief complaints in primary care. However, dermatologists are consistently in short supply, particularly in Chinese rural areas. Artificial intelligence tools can assist in diagnosing skin disorders from images, however the database for Chinese population is very limited, and it’s also non-trivial to directly apply the datasets built upon the US or EU population.

Objective:

To establish a dataset for artificial intelligence based on Chinese population, and present an initial study on six common skin diseases.

Methods:

Each image is captured with digital cameras or smartphones and verified by at least 3 experienced dermatologists and corresponding pathology information, and finally formed the Xiangya-Derm database. Based on this database, we conducted artificial intelligence-assisted classification research on 6 common skin diseases and then proposed a network called SkinNet. SkinNet applied a two-step strategy to identify skin diseases. Firstly, given an input image, we segment the regions of the skin lesion; Secondly, we introduce an information fusion block to combine the output of all segmented regions. We compare the performance with 31 dermatologists of varied experiences.

Results:

Xiangya-Derm, as a new database which consists of over 150,000 clinical images of 571 different skin diseases from Chinese population. It is known to be the largest and most abundant dermatological dataset of the Chinese. The artificial intelligence–based six-classification achieves the top-3 accuracy of 84.77%, which outperforms the average accuracy of dermatologists (78.15%).

Conclusions:

Xiangya-Derm, a new and the largest database for the Chinese population was formed and the accuracy of classification of six common skin conditions based on Xiangya-Derm is comparable to that of dermatologists.