Research On Assistant Diagnosis Of Diabetic Retinopathy Based On Image Processing

Diabetic retinopathy(DR)is a very common complication of diabetes.It has become one of the most important diseases leading to blindness in China,and timely diagnosis and treatment can effectively avoid its harm to vision.However,due to the relative shortage of medical resources,DR screening and prevention work has become extremely difficult,which has caused a large number of DR patients to miss the best time for treatment,which ultimately leads to visual impairment or even blindness.To this end,the use of digital image processing and machine learning to achieve automatic diagnosis of diabetic retinopathy is of great significance for the realization of large-scale fundus screening.The thesis focuses on the segmentation of fundus image and the staging diagnosis of diabetic retinopathy.The specific content and innovation results are as follows:(1)in order to remove the interference of vascular network on lesion extraction,this paper proposes a vascular network segmentation algorithm based on digital morphology.The rough segmentation of the vascular network was obtained by the base cap transformation,and then the fine segmentation of the vascular network was completed based on the morphological reconstruction theory.(2)in order to remove the interference of optic disc for lesion extraction,this paper proposed a fast optic disc localization method based on sliding window,which can complete the localization of optic disc in a very short time.(3)in order to realize the segmentation and extraction of hard exudate,this paper proposes a hard exudate segmentation algorithm based on image fusion based on the characteristics of the exudate.Compared with other algorithms in the same field,this algorithm has the characteristics of specificity,high accuracy and short time consumption.(4)for staging diagnosis of diabetic retinopathy,this paper puts forward a kind of based on support vector machine(SVM)auxiliary diagnosis algorithms of diabetic retinopathy in installment,starting from the pathological characteristics of diabetic retinopathy to extract the feature vector,the classifier training and testing,test suitethe results show that the model for simplex diabetic retinopathy middle diagnosis effect is good.