Abstract

Info

Author(s):: ZHANG Ying1; ZHAO Qiyang1; XI Qun2; 1. School of Information Engineering, Gansu University of Chinese Medicine, Lanzhou 730000, China; 2. Information Center, Lanzhou University Second Hospital, Lanzhou 730000, China

Keywords:: diabetic retinopathy; deep learning; deep residual shrinkage network; pyramid split attention module

Abstract:: A converged network based on deep learning is proposed to realize the efficient and accurate diagnosis of diabetic retinopathy. Both data augmentation technology and generative adversarial network are used to augment the fundus images in EyePACS dataset for effectively addressing the problem of uneven classification of fundus images. The proposed model uses Inception-Resnet-V2 as the main network, and incorporates deep residual shrinkage network and pyramid split attention module for effectively filtering out the irrelevant information in the feature learning process and focusing on the lesion information, thereby improving the network’s ability to capture important features. Experimental results show that the optimized model achieves accuracy, recall, specificity, sensitivity, and F1 score of 0.951, 0.950, 0.990, 0.950, and 0.950, respectively, without the need to specify lesion characteristics in advance, demonstrating its superiority in evaluation indicators.