A Novel Approach for Diabetic Retinopathy Screening Using Asymmetric Deep Learning Features

Automatic screening of diabetic retinopathy (DR) is a well-identified area research in the domain computer vision. It challenging due to structural complexity and marginal contrast difference between retinal vessels background fundus image. As bright lesions are prominent green channel, we applied contrast-limited adaptive histogram equalization (CLAHE) on channel for image enhancement. This work proposes novel technique using an asymmetric deep learning feature. The features extracted U-Net segmentation optic disc blood vessels. Then convolutional neural network (CNN) with support vector machine (SVM) used DR classification. classified into four classes, i.e., normal, microaneurysms, hemorrhages, exudates. proposed method tested two publicly available datasets, APTOS MESSIDOR. accuracy achieved non-diabetic detection 98.6% 91.9% MESSIDOR respectively. accuracies exudate these datasets 96.9% 98.3%, system improved precise segmentation.