Research On Detection Of Microaneurysms In Diabetic Retinopathy Disease From Retinal Fundus Images

At present,one of the difficulties in the process of diagnosis and treatment of diabetic retinopathy is how to safely and efficiently inspect retinal images and identificate retinopathy in a timely manner so that as soon as possible to take corresponding measures to prevent the deterioration.In this paper,an in-depth study of the detection methods of microaneurysms in diabetic rctinopathy has been carried out.The main contents are as follows:(1)This paper introduces the research background and the basic theory of microaneurysm detection.Firstly,this paper summarizes the research background and significance of microaneurysm detection technology,and points out the existing problems in this field.Then,the pretreatment technology of the retinal image and the random forest algorithm for classificating the features of the microaneurysm areas are introduced.(2)This paper implements the preprocessing and candidate segmentation methods of microaneurysm detection.Firstly,in view of the retinal image of uneven illumination and various color,this paper adoptes the method of digital morphology processing and improves the retinal image visual effect,which is advantageous to the subsequent microaneurysm region segmentation.Then,for the fact that retinal microaneurysm detection is vulnerable to image background,blood vessels and optic disc,this paper adopts digital morphology processing,hough transform and GrowCut algorithm,improving the accuracy of the microaneurysm candidate region segmentation.(3)This paper implements the microaneurysm detection algorithm based on weighted random forest.First of all,for microaneurysm detection does not consider the texture feature,this paper joins the gray level co-occurrence matrix feature in feature extraction,carries on the quantitative and qualitative analysis in the experiment,and verifies the rationality of the gray level co-occurrence matrix feature.At the same time,the contrast experiment shows that the features of gray level co-occurrence matrix have indeed improved the accuracy of microaneurysm detection.Then,in view of the low accuracy of the traditional random forest microaneurysm detection algorithm,on the basis of the traditional random forest model,this paper implements a weighted random forest algorithm,namely before the vote,the classification in advance will vote for each decision tree and change the vote ability into the weight of corresponding to its training accuracy.The result of the algorithm in ROC database is 77.01%sensitivity and 90.75%specificity.In addition,experimental result shows that the test results of weighted random forest are better than those of traditional random forest and several typical classifiers.(4)This paper designs and implements the GUI system of microaneurysm detection,and summarizes and prospectes the microaneurysm detection algorithm.