Application of a deep learning system for detecting prediabetes based on eye image and life style

Diabetes mellitus (DM) is one of the most graving public health challenges worldwide that brings enormous harm and burden to patients and societies. Due to its complicated diagnostic methods and limited screening manners, the detection rate of patients with early diabetes mellitus is very low worldwide. As the deep learning algorithms that feature using image recognition to extract characteristics mature, early artificial intelligence(AI)-based diabetes mellitus screening model has been initially established. However, external clinical validation in population of different ethnicities is still lacking. In this study, we hope to use the color fundus images and lifestyle-related variables (such as smoking, drinking, and exercising) in the database to validate the established AI-based model. The results will prove the validity of the model, which will help improve not only the screening detection rate of early diabetes mellitus and other eye diseases, but also the screening coverage of diabetes mellitus, thus reducing the labor cost of screening.