Ophthalmic Diseases Detection using CNN: A Comparative Analysis of ResNet and VGG Architectures

Ocular illnesses present a considerable risk to worldwide public health, frequently resulting in visual impairment or blindness if not identified and addressed appropriately. This study focuses on the essential task of automated identification of eye diseases via convolutional neural network (CNN) models. With the increasing availability of digital retinal imaging data, there is a critical need for precise and efficient diagnostic tools to aid healthcare personnel in identifying common eye diseases, including diabetic retinopathy, cataracts, glaucoma, and normal eye conditions. The majority of eye disease diagnoses now occur through manual examination by ophthalmologists, which can be laborious and prone to inter-observer variability. By leveraging CNN architectures like VGG19, VGG16, ResNet18, and ResNet50, we propose a robust framework for automated eye disease detection. Our methods involve preprocessing retinal images, training CNN models on labeled datasets, and evaluating their performance on unseen data. Our results showcase high classification accuracy, sensitivity, and specificity across all disease categories. Specifically, our models achieved an accuracy of 92% for VGG19, 91% for VGG16, 87% for ResNet18, and 85% for ResNet50. The findings offer a promising solution to streamline the diagnostic process and improve access to timely eye care services. Integrating this CNN-based system into clinical workflows can expedite diagnosis, enable early intervention, and prevent vision loss, contributing to advancements in computer-aided diagnosis systems and enhancing efficiency in medical imaging analysis.