ML and Deep Learning for Diabetic Retinopathy Severity Detection

Diabetic retinopathy (DR) is known to be the major entanglement as a side-effect of diabetes. This disease majorly causes certain amount of injury of blood vessels in the area of retina. The extent of wound varies in magnitude that concludes the severeness of the disease. If DR is not predicted or treated on time, it may cause permanent loss in vision. Most recent studies prove that, computer aided diagnosis of retinal fundus images provides most promising and reliable outcomes. This paper compares the efficacy of traditional machine learning (ML) and deep learning (DL) architectures, including k-nearest neighbour (kNN) ML algorithm, vision transformers (ViT), ResNet50 and VGG16 DL algorithms for classification of diabetic retinopathy based on the severity of the disease. The paper has utilized a data set comprising five distinct classes ranging from mild to severe stages of the disease. Remarkably, the ViT emerges as the top performer surpassing other methods. The study offers promising avenues for the development of more accurate and efficient diagnostic tools for diabetic retinopathy assessment.