A Comparative Analysis on Diabetic Retinopathy using Deep Learning and Nature based Optimization Algorithms

Abstract

Diabetes, often known as diabetes mellitus (DM), is a metabolic disease caused due to high blood sugar levels. Diabetes results in Diabetic Retinopathy (DR), an eye condition that over time significantly impairs vision. Acute retinal symptoms can include micro aneurysms, swollen blood vessels, fluid leaks, exudates, and haemorrhages. Today's medical science relies heavily on images to diagnose patients correctly. In recent years, a model for computer vision has been effectively trained utilizing deep neural networks and nature-based optimization methods, exceeding earlier neural network models in terms of accuracy. By utilizing a hybrid version to assist ophthalmologists in doing these assessments, this study aims to enhance their capacity to diagnose DR. The IDRiD collection of retinal pictures, which is openly available, was used in this study. In the suggested model, features were extracted from the input photos using a deep learning model, specifically InceptionV3 and DenseNet-169, and classification was performed using the optimization algorithms PSO, ACO, and BEE. The fully connected layer's output weights were optimized using weights that were learned using ImageNet data to provide suitable features. This work demonstrates that a pre-trained system combined with a metaheuristic algorithms may reliably diagnose DR.