Classification of Diabetic Retinopathy Severity Using Deep Learning Techniques on Retinal Images

Abstract

In order to diagnose DR by utilizing the grayscale intensity and texture information extracted from the fundus image, deep learning approach is used. The writers came up with the strategy on their own. The APTOS 2019 Blindness Detection (APTOS 2019 BD) dataset is employed. extensively during the course of this investigation. A method powered by deep learning was used to locate and recover the photos. We use a large number of image processing methods, as well as two ways to feature extraction, and one method for feature selection. Because of this, the F-measure is 92.3% (0.5%), and the classification accuracy is 93.2% (with an error margin of 0.32%). Durability and dependability of the approach that was proposed have been shown by performance-related criteria for the operation. There is a possibility that classification performance might be improved using a weighted ensemble model that makes use referring to the models EfficientNet-B0, B5, and B7.This research highlights the need of extra diagnostic procedures being performed by medical experts in order to avoid incorrect diagnoses as well as false positives.