Study On Image Grading Method Of Diabetic Retinopathy Based On Deep Learning

Diabetic Retinopathy(DR)has become the leading cause of the increasing number of blind patients worldwide,while high-definition color fundus images have brought great convenience for DR diagnosis.However,manual diagnosis takes time and effort,and different doctors may make different diagnoses.At present,intelligent grading based on deep learning has become a hotspot in DR intelligent diagnosis.Although the DR intelligent classification models based on convolutional neural network(CNN)have achieved good results,these models pay more attention to the extraction of deep features,and do not consider the intrinsic relations between features,the characteristics of the DR image itself and relationships between different levels,etc.This paper proposes two intelligent classification models by referring to the powerful relationship capturing ability of graph neural network(GNN)and the powerful detail capturing ability of the capsule network.The main work and innovation achievements are as follows:(1)A DR images classification model based on GNN is proposed.In order to overcome the problem that the feature difference of DR image is small and it is difficult to distinguish between adjacent levels,it is considered that the relationship between the deep features of DR images contains important classification information.The model is composed of two cascaded networks,CNN is used to extract the deep features of the DR images,and GNN is used to further capture the relationship between the deep features of CNN.At the same time,considering the contribution ratio of the two networks to the final result,the model proposes to use the adaptive weight mechanism to complete the fusion of the two networks,and gives the classification results of the whole network.Finally,the performance of the model is evaluated on two datasets and good results are obtained,which proves the superiority of the model.(2)A DR images classification model composed of two cascaded based on capsule network and GNN is proposed.First,with the detail capture capability of capsule network,the pooling layer of CNN is replaced by capsule network to extract the deep detail features of DR images.Secondly,considering the small difference of DR images between adjacent levels,which is easy to be confused,the GNN is used to capture the relationship between DR levels.Finally,the output of the two networks is fused through the adaptive weight,and give the classification results of the whole network.The proposed model is evaluated on two datasets respectively,and good results are obtained,which further demonstrates the superiority of the proposed method.The above two models are based on the characteristics of DR images.The first model focuses on considering the relationship between deep features of DR images contains important hierarchical information,while the second model focuses on considering the relationship between levels of DR images contains important classification information.The ultimate goal is to improve the classification performance of DR images.