Retina Disease Classification using Deep Learning

Abstract

The most common causes of vision impairment worldwide include retinal conditions like Drusen, Diabetic Macular Oedema, and Choroidal Neovascularization (CNV). Early detection and treatment are essential to preventing irreversible harm, yet traditional diagnostic methods are expensive and necessitate specialized knowledge, making them occasionally unavailable in under-resourced locations. The objective of this study is to develop a scalable and reasonably priced early detection tool that automatically classifies retinal abnormalities from grayscale fundus images using deep learning techniques. In this study, I employed transfer learning with pre-trained models like EfficientNetB4, ResNet50, and VGG16 to classify retinal illnesses using a carefully chosen dataset of greyscale fundus images. The models were evaluated using a number of performance metrics, including F1-score, recall, accuracy, and precision. The results showed that the transfer learning models, particularly EfficientNetB4 and ResNet50, outperformed VGG16, with EfficientNetB4 achieving the best accuracy. The findings demonstrate that deep learning models may accurately classify retinal diseases, especially those that incorporate transfer learning. These models may find application in telemedicine, automated screening systems, and perhaps improving accessibility to eye disease diagnosis. Future studies will focus on expanding the dataset, refining the models for real-time applications, and exploring more complex structures in order to enhance classification performance.