Research On Analysis Algorithm Of Diabetic Retinal Images Based On Salp Algorithm

In recent years,the number of diabetic patients has been increasing year by year,and most patients with diabetes mellitus are accompanied by retinopathy in the advanced stage.However,the vision of patients will not significantly decline before the lesions invade the macular area,so most patients miss the optimal treatment stage due to the difficulty in self-examination of the initial symptoms of retinopathy.Artificial examination of retinopathy images will often result in false examination due to factors such as poor quality of the image,too many lesions,too small lesions,etc.Therefore,this paper studied the segmentation of diabetic retinopathy RGB images and fluorescence fundus angiography images,and the identification of microaneurysms,respectively.The main research contents are as follows:To solve the RGB blood vessel image segmentation problem in diabetic retinas,an RGB blood vessel image segmentation method based on improved PSO-SALP algorithm was proposed.First,matched filter and high-low cap transform were used to enhance the details of the terminal vessels.Secondly,particle swarm optimization(PSO)is used to improve the salp algorithm according to the discrete criterion.Then,the RGB vascular image is segmented by the maximum between-class variance criterion to comprehensively evaluate the image segmentation performance.Finally,the obtained images are combined with Frangi filtering and morphological reconstruction images.The results showed that the accuracy and specificity of the method were 96.3% and 98.5%,respectively.The diabetic retinal microaneurysm of RGB image microaneurysm recognition,attention to design a fusion mechanism of Residual Attention Unet network,realize the RGB image in microaneurysm detection.Aiming at the low proportion of positive cases in medical images,the Residual Attention Unet network integrating attention mechanism was designed on the basis of Res Unet34,and the extended ROC data set was used to train the network model.Adjust the weight of positive cases in Dice-Loss to neutralize the influence of low proportion of positive cases in medical images;Finally,a comparative analysis experiment was carried out with other classical models,and the results showed that the designed RA-Unet model could correctly identify microaneurysm in RGB images of diabetic retinas.To solve the problem of microaneurysm recognition in fluorescence fundus angiography images of diabetic retinas,a clustering based microaneurysm detection method was proposed.As microaneurysm is a highlighted small point in fluorescence fundus angiography image,it is easy to regard microaneurysm as noise elimination in image preprocessing or segmentation.Therefore,a method combining connected domain and multi-valued salp algorithm is proposed in this paper to obtain binary images containing microaneurysm,and candidate microaneurysm sets are obtained after morphological operation.The improved Multi-Salp algorithm K-means algorithm is used to segment the candidate microaneurysm set with multi-feature clustering.The Radon transform of the candidate microaneurysm set is performed with 18-angle Radon transform and its standard deviation is obtained.Finally,microaneurysm was located by clustering segmentation standard deviation value.The results show that the average precision of the proposed method can reach93.24%.