Diabetic Retinopathy Screening Using Image Processing

Abstract

Diabetic retinopathy, a grave consequence of diabetes mellitus, has emerged as the leading cause of visual impairment worldwide. This ocular condition arises from the deterioration of blood vessels situated behind the retina and progresses insidiously, ultimately leading to blindness. Early detection is paramount in mitigating vision loss among afflicted individuals. In this study, we propose three distinct approaches Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and Residual Networks (ResNets) for the accurate detection of diabetic retinopathy. Our aim is to determine the most effective model for this purpose, thereby improving screening efficiency. Utilizing a pre-processed dataset sourced from Kaggle, we conducted comprehensive experiments to evaluate the performance of each model. This curated dataset was instrumental in optimizing the classification algorithms. Our findings reveal notable disparities in the performance of these models. Through meticulous testing and validation, we sought to identify the model exhibiting the highest accuracy in diabetic retinopathy detection. Leveraging a dataset comprising 2750 retinal images, our experiments yielded accuracy values of 68% for SVM, 74% for CNN, and 63% for ResNet.