Automatic Classification of Diabetic and Hypertension Fundus Camera Retinal Images Using Deep Learning

Background and Objectives: Various systemic diseases cause changes in the retina. In India, diabetic retinopathy (DR) is the leading cause of blindness, while hypertension retinopathy (HyT) alleviates various disease complications. Fundus image analysis, traditionally conducted by an ophthalmologist, identifies the stages of Diabetic Retinopathy (DR) and Hypertensive Retinopathy (HyT) symptoms. However, computer-aided systems have proven to be great assistive technology providing fast and cost-effective solutions for the first-level screening of these diseases. Material and Methods: The classification of fundus images into diabetic and hypertensive retinopathy disease is proposed in this work. To assess whether images belong to DR (at different stages) or HyT, we propose two architectures: Convolutional Neural Network (CNN) and Transfer Learning with VGG-16. We utilized these models to classify diabetic images (at the different stages: from diabetic mild stage 1 (DS1) to diabetic severe stage 4 (DS4)) and hypertensive images. We have tested the system performance on images taken from various publicly available datasets. Results: The proposed system’s performance was evaluated using metrics such as accuracy, precision, recall, and F-score. It was observed that F-score, precision, and recall is very high using CNN for both the diabetic and hypertensive class. The same is true when the classification was performed using VGG16. It has been seen that the VGG-16 model performs better for classifying hypertensive retinal images from severe diabetic stage retinal images. The designed system is reliable and robust enough to classify the different diabetic stages from hypertensive fundus images.