Segmentation of Diabetic Retinopathy Fundus Images Using Residual U-Net with the Sigmoid Function based Dipper Throated Optimization Algorithm

The identification of retinal microaneurysms and exudates in fundus images is essential for the early diagnosis of Diabetic Retinopathy (DR).However, conventional approaches face challenges due to dimensional complexity and limited differentiation of microaneurysms in fundus images.Hence, this research proposes a Residual U-Net for the segmentation of red lesions in DR fundus images.The incorporation of residual connections in the U-Net enable it to bypass unnecessary layers of the network, mitigating the issue of vanishing gradients in deeper networks.Three standard colour fundus image datasets namely, E-Ophtha, DiaRetDB1 and Messidor are used to evaluate the effectiveness of the proposed approach.In the pre-processing step, fundus image resizing and min-max normalization techniques are utilized to enhance the quality of the input images.Then, Local Binary Patterns are extracted to retain only the relevant features from the segmented data.Furthermore, the Sigmoid function-based Dipper Throated Optimization (SDTO) approach is used for feature selection, while a Convolutional Neural Network (CNN) is employed for the classification of DR into binary classes.The experimental outcomes demonstrate that the proposed Residual U-Net with the SDTO approach achieves a better accuracy of 99.34% and 99.93% respectively on E-Ophtha and DiaRetDB1 datasets, thereby facilitating earlier detection and DR diagnosis than the existing methods like the CNN and U-Net.