A Fundus Image Myopia Diagnosis Model Based on Homogeneous Multimodal Feature Fusion

High myopia is one of the leading causes of fundus diseases. If it can be found and treated in time, the risk of fundus lesions in children's growth will be reduced, and the growth rate of patients with visual disabilities will be effectively controlled. Vascular, as one of the significant features in fundus images, are often used as additional elements to help ophthalmologists diagnose. Hence, in this paper, based on the idea of homogeneous multimodality, we design a neural network model with two branches that simultaneously processing the vascular feature image and original fundus image therefore to automatic detect high myopia based on the fundus images, and hope to make a great difference in clinical practice. Extensive comparative experiments were conducted between our method and other general classification models through a private retinal fundus data set. The results show that our method achieves the best performance of 93.4% in accuracy.