Diabetic Retinopathy Study Convolution Based On Neural Network

Diabetic Retinopathy,as a kind of microvascular disease,is often caused by diabetes.In the early stage of the disease,the symptoms are not obvious,and the condition deteriorates rapidly.It can cause vision loss in the eyes,and it can cause blindness.Its clinical manifestations focus on the vascular disease of the fundus.At present,in clinical practice,professional ophthalmologists rely on their own experience to find potential diseased areas from the retinal images of the fundus,analyze and screen them,and thus can evaluate and diagnose the patient’s condition.However,on the one hand,due to excessive reliance on the clinical experience of ophthalmologists,diagnosis takes a long time and is prone to missed and misdiagnosed situations;on the other hand,some areas are limited by medical conditions,which makes it impossible for many patients to get the diagnosis and treatment in time.It can cause serious consequences.Therefore,under the effect of computer-assisted technology,through traditional image processing and convolutional neural network methods,the fundus retinal image is processed and analyzed,so as to achieve the role of auxiliary clinical treatment.This paper combines traditional image processing technology and convolutional neural network to study the fundus retinal blood vessel segmentation and DR classification.The specific research content is as follows:(1)In order to make the images in the data set adapt to the task,fully consider the multiple public data sets used and the characteristics of the research task to be carried out,and use a variety of methods to process the data set,including channel selection,contrast adjustment,Region of interest extraction,normalization and data amplification,etc.Among them,data amplification adopts many methods,such as up and down,left and right,translation,rotation,and smoothing.(2)In order to solve the problem that the fundus images are easily affected by lesions and the segmentation of vessels is intermittent in the task of retinal vessel segmentation,a fundus vessel segmentation method based on improved matched filter and PCNN is proposed.On the one hand,Hessian is used to guide the matching filter angle and multi-scale operations are performed to enhance the details of blood vessels.On the other hand,morphological methods are used to reduce the impact of lesions.Finally,PCNN is combined with the idea of regional growth to achieve precise segmentation of retinal vessels.On the DRIVE and STARE data sets,the accuracy reached 0.9462 and 9460,and the AUC reached 0.9638 and 0.9605,respectively.(3)This paper presents a classification algorithm for diabetic retinopathy based on the A-Inception-Resnet-v2 model.The algorithm uses the Inception structure with high depth and width and the residual structure to extract retinal image features faster and effectively.The introduction of the attention mechanism allows the network to pay more attention to the lesion features when extracting features.The A-Inception-Resnet-v2 classification model provides a solution to the problem that the difference between the categories of retinal image lesions is small,which makes it impossible to subdivide.This has a positive impact on the classification of DR.The classification accuracy reached 92.6% on Eye Pacs dataset,and the quadratic weighted Kappa value reached 80%.