The Research On Automatic Classification Of Diabetic Fundus Images Based On Deep Learning

In recent years,with the rapid development of artificial intelligence technology,it has been widely concerned by all walks of life.The combination of artificial intelligence technology and modern medical equipment has become the trend of modern medical technology development.Retinal fundus image carries pathological information which plays an important role in the diagnosis of diabetes mellitus.In most cases,the traditional processing of retinal fundus image is limited by doctors’experience,and the efficiency is poor.In this paper,an automatic classification system of diabetic fundus image based on web is constructed by making full use of artificial intelligence technology.The research in this paper mainly involves the following aspects:1.In the extraction stage of retinal fundus feature images,the improved U-net network is used to extract the feature images including blood vessels,bleeding points and exudation from the original retinal fundus images.The results of the three feature sub-images are fused to prepare for the input of the feature images of the classification CNN network.2.Realization of retinal fundus image classification network.In this paper,the classical Inception-ResNet-v2 network and Xception network are selected as the backbone network to carry out the comparative experiments.A two-tower classification network model is designed in this paper,and the classification network is trained with parameters which were trained on ImageNet.3.The feature vectors extracted by the network are weighted by genetic algorithm and classified by the support vector machine.4.The realization of visual page of diabetic retinopathy fundus image detection based on web,using Node.js as the development language,combined with the results of automatic classification of diabetes grade are displayed visually,and the output of automatic diagnosis report.The automatic classification of diabetic retinopathy level is realized on the browser side.In this paper,deep network is used to extract feature images and automatic classification of diabetic retinal fundus images.We strive to improve the accuracy of automatic classification of diabetic retinopathy by fusing the original image with supervised feature images.At the same time,the visualization of classification process and classification results of diabetic retinopathy images are realized.It greatly improves the diagnostic efficiency of doctors and promotes the practical process of landing research results to a certain extent.