In order to improve the early identification ability of high-risk people with type 2 diabetes mellitus, a risk warning model based on multi-modal data fusion was constructed. In this study, the clinical indicators, continuous monitoring signals, lifestyle information and fundus image data of 3126 subjects were integrated. After preprocessing, single-modal feature extraction, dynamic weighted fusion and gated restructuring, the risk warning framework of type 2 diabetes mellitus was established. The results show that the accuracy, F1 value and AUC of the proposed model on the test set reach 0.934, 0.906 and 0.978, respectively, which is better than that of the traditional machine learning model, the single-modal deep model and the conventional early fusion model. This method can more fully characterize the individual metabolic risk characteristics and provide a feasible computational support for the early screening and early warning of type 2 diabetes.
