A Machine Learning Approach for the Diagnosis of Diabetic Retinopathy

Abstract

Diabetic retinopathy (DR) continues to be a major cause of preventable blindness globally, with early diagnosis a critical factor in the prevention of loss of vision. Diabetic Retinopathy (DR) is one of the main causes of avoidable blindness globally, and its diagnosis is most crucial for intervention. In this work, we propose an automated DR classification system using the Messidor dataset of fundus retinal images. Our approach primarily uses Convolutional Neural Networks (CNNs) for feature extraction and DR stage classification, given their success in medical imaging. To address the scarcity of annotated medical data, data augmentation methods such as random flips, rotations, and brightness changes are applied to improve the model's generalisation. In addition, transfer learning using pretrained CNN models (e.g., ResNet and EfficientNet from the timm library) is employed to accelerate convergence and improve accuracy. For thorough examination, the dataset is split into training, validation, and test subsets, and model efficiency is quantified through metrics such as accuracy, precision, recall, F1-score, and AUC.