Enhanced convolution neural network and improved SVM to detect and classify diabetic retinopathy

Blood vessels in the retina become damaged as a result of diabetes. If it is not treated early, then it leads to Diabetic retinopathy (DR), which is a significant reason for blindness. Clinical tests are needed to detect and classify the stages, but these are very expensive and time-consuming. This paper proposes an enhanced Gaussian distribution-based image enhancement and segmentation to detect diabetic retinopathy. Medical images have high noise and artefacts, making it difficult to recognize the details with the normal eye and very tough to process the image further. The existing approaches to enhance medical images are very complex and time-consuming. We proposed a more effectiveenhanced Gaussian distribution approach to overcome this limitation. The lesions' inconsistent size and noise in the fundus image make it challenging to segment the lesion. In this study, an improved regularization function in the convolution neural network(CNN) was applied to the retina images to detect the lesions automatically and reduce the overfitting, reducing loss. An improved Support Vector Machine(ISVM) classifies the segmented fundus images. The proposed improved CNN model is compared with the previously carried out studies, and the results prove that the improved CNN model is efficient in diagnosing diabetic retinopathy.