A Deep Learning–Based Approach for Detecting Normal and Abnormal Diabetic Retinopathy

Abstract

Diabetes can gradually sneak up on you to impair your eyesight. When retina measurements are not taken often, small but significant changes in retina are missed out. If detected early enough, permanent vision loss can be stopped. Nevertheless, a great deal of time is wasted reviewing the retinal scans manually. Different reviewers may arrive at different conclusion about a decision. Currently, trained specialists examine the images and do this type of work. The Current Approach Seems Unfeasible For At Scale Implementation Of The Solution. It is a deep-learning-based system that makes an effort to distinguish between normal and abnormal status of retina without any kind of human involvement. To design a system to capture images spatially, morphologically and temporally, novel deep networks architectures have been proposed. The architectures utilize CNN, RNN, bi-directional RNN, auto encoders and VGG-16 network. Other learning techniques such as Decision Forest are used to compare performance assessment Publicly available Diabetic Retinopathy datasets yield experimental results. The suggested method attains excellent accuracy with solid generalization across various. Such images are good quality and effective for early DR detection. An automatic system. Ophthalmologists may perform screenings quickly and routinely to lessen heavy work. This. The structure enhances strength and feature representation and ability to separate normal. using an amalgamation of models to create an abnormal retinal image.