Lightweight Vision Transformer Network for Diabetic Retinopathy Classification Through Data Balancing Approach

Diabetic Retinopathy (DR) is an irreparable fundus retinopathy which primarily affects people with persistent diabetes and is one of the major causes of blindness worldwide. Early detection and accurate classification of DR is essential in preventing vision loss. Due to insufficient representation of lesion information, the conventional deep learning (DL) techniques for DR grading are frequently time-consuming and produce inferior detection performance. To overcome these challenges, in this study, we propose a lightweight vision transformer network and classify DR into five different categories: No DR, Mild DR, Moderate DR, Severe DR, and Proliferative DR. Since the dataset showed innate class imbalance, we executed data balancing techniques to produce a robust model training and evaluation. Following a comprehensive assessment of our model’s functionality, we found a significant accuracy rate of 96% as well as improved performance metrics. Additionally, our network showed impressive kappa score of 97.50%. Our findings demonstrate that balanced data can substantially improve classification performance, confirming the importance of the lightweight vision transformer approach for automated, extensive DR screening in clinical settings.