Proposal for the Development of an Easily Accessible System for the Early Detection of Diabetic Retinopathy for Non-Experts

Abstract

Diabetic retinopathy is the most common condition among patients with diabetes mellitus and the main cause of vision loss. This is because DR can evolve for years without its carrier detecting it. However, in a DR exam, image capture is a manual process that can take a long time. Advances in deep learning algorithms have made it possible to learn the most predictive features of DR directly from fundus images obtained from patients with diabetes. These images are used to train deep learning models that can identify patterns of DR in new patients. In turn, existing models have been shown to perform as well as or better than a specialist in the area. The present study reviews the state of the art of models used for DR detection and establishes a possible theoretical proposal for the creation of an imaging system to obtain fundus captures. In addition, the use of existing deep learning models to process the images and determine whether the patient may have DR.