RetinFormer : Augmented Vision Transformer based diabetic and Hypertensive Retinopathy Classification

Diabetic and hypertensive retinopathy are severe complications of patients who suffer with diabetes and hypertension that can lead to blindness if not diagnosed early. Current diagnostic methods are time-consuming and require specialized expertise, which necessitates an automatic classification of retinal images to reduce diagnostic time and improve healthcare accessibility. To address this, we propose a new augmented vision transformer (ViT) based diabetic and hypertensive retinopathy classification method, termed RetinFormer. By utilizing a ViT with pooling layers, the model efficiently extracts features from retinal images for classification into diabetic retinopathy, hypertensive retinopathy, or normal images. The proposed method is evaluated on three publicly available datasets. Data augmentation is employed to increase sample size and improve model generalization. Experimental results show that the RetinFormer achieves a peak of 92.71 % accuracy with an average F1-Score of 0.9270, outperforming existing state-of-the-art models.