Study On Detection Of Diabetic Fundus Exudates And Taging Of Diabetic Retinopathy

In recent years,with the improvement of living standards and changes in eating habits,the incidence of diabetes has increased year by year,and it has become a chronic disease that has a widespread impact on people’s health.Diabetic retinopathy is an extremely common and serious complication of diabetes,and it has become one of the main causes of blindness among adults in my country.Therefore,timely diagnosis and treatment can effectively avoid harm to people’s eyesight.At present,screening for diabetic retinopathy is mainly performed by ophthalmologists manually inspecting patients’ fundus images.However,due to complex fundus images,long manual inspection time,and low efficiency,many patients cannot receive timely and effective treatment.Therefore,computer-assisted image processing is used to realize automatic detection of pathological changes in fundus images,so as to improve the diagnosis efficiency of doctors.This paper mainly proposes a method for detecting hard exudate based on SVM based on retinal blood vessel segmentation and optic disc positioning.Based on the detection of hard exudate,by extracting different characteristics of hard exudate,based on support The vector machine realizes the research on the staging of diabetic lesions.The main research contents are as follows:(1)Research and implement a semi-trained generalized linear model of retinal blood vessel segmentation method.The method of CLAHE algorithm and mean filter difference processing to get the preprocessed image is introduced,and combined with the features of the lower half of the preprocessed image based on Gabor feature extraction,a generalized linear model is constructed to realize the blood vessel segmentation of the entire image.On the basis of blood vessel segmentation,an algorithm for optic disc positioning and segmentation is implemented based on features such as optic disc appearance.First,the pre-processed image is set to a certain threshold for binary segmentation to obtain the initial result of the optic disc positioning;then,the connected area is extracted,and the interference of the edge area of the optic disc is filtered out to realize the accurate positioning of the optic disc;finally,the precise positioning is performed The result is segmented by Hough transform.The average value of the performance indicators of the blood vessel segmentation algorithm in this paper:accuracy rate of 0.9442,sensitivity of 0.8879,specificity of 0.9353,and AUC of 0.9677.Various performance indicators verify that the algorithm has good segmentation performance and applicability.(2)According to the color and brightness of retinal fundus exudates in diabetic patients,research and implementation of a hard exudate detection algorithm based on support vector machine.First,select the M-channel image in the CMY color space as the background image for preprocessing;second,perform threshold segmentation on the preprocessed image,and combine the optic disc segmentation algorithm proposed in Chapter 2 to eliminate the optic disc and determine the candidate area of the exudate;and finally,Extract the different characteristics of the exudate in the candidate area,construct and train the training model on the basis of the support vector machine to realize the detection of exudate.The detection algorithm proposed in this paper is randomly tested in the DIARETDB1 library,and the evaluation criteria based on the lesion area are used to measure the algorithm in this paper.The experimental results have obtained a sensitivity of 93.92%,a specificity of 89.2%,an accuracy of 94.7%,and a prediction rate of 95.4%.Good performance indicators,and comparing each performance indicator with other exudate detection algorithms,verify the good applicability of this algorithm.(3)Based on the knowledge research on the staging of diabetic retinopathy and the detection and segmentation of hard exudates in Chapter 3.The support vector machine realizes the function of staging and classification of diabetic retinopathy.First,extract the appropriate features of the hard exudate,select training samples and test samples from the data set,and train the classifier based on the training set based on the support vector machine classification theory according to the staging standard of diabetic retinopathy,and test it on the test set.The test results show that for normal fundus images,the classifier can accurately judge and recognize,and the accuracy P value reaches 100%.For early NPDR and mid-term NPDR,the classifier has a recognition accuracy of 90%,and the accuracy is high,but for Severe NPDR,this classifier has low recognition accuracy.It shows that the classification device has a high accuracy in the early stage diagnosis and recognition of diabetic retinopathy,and can generally meet the requirements for early stage diagnosis of diabetic retinopathy.