Deep Learning Methods For Classifying Multiple Retinal Diseases Using Fundus Images

Abstract

Retinal diseases such as Diabetic Retinopathy (DR), Age-Related Macular Degeneration (ARMD), Optic Disc Cupping (ODC), Glaucoma, and Myopia are leading causes of vision impairment and blindness globally. Early detection through automated screening using deep learning is crucial for effective treatment and vision preservation. This study investigates the application of deep learning methods for the multi-class classification of retinal diseases using fundus images. Seven models—VGG16, ResNet50, EfficientNetB7, DenseNet201, EfficientNetB4, ResNet152 and a Customized CNN—were evaluated on a balanced dataset of 600 images divided into six classes. Each model was assessed using performance metrics such as accuracy, sensitivity, specificity, precision, and F1-score. Among the evaluated models, the Customized CNN achieved the highest overall accuracy (91.37%) and demonstrated superior classification performance across most disease categories. The findings emphasize the potential of tailored CNN architectures in clinical screening applications and support the development of user-friendly diagnostic tools for early disease identification.