Ocular Disease Detection via Deep Learning Models

There has always been a problem for clinicians to identify eye diseases using fundus images early enough. Ocular disease diagnosis by hand is an expensive task, prone to errors, and difficult. Automated computer system for detecting ocular diseases is needed to identify different eye illnesses utilizing fundus pictures. This research introduces a deep learning-based approach for precise ocular detection. With the advancements in deep learning algorithms, image classification has significantly improved, making such a system more efficient and reliable. For this, we classified the ODIR dataset, which includes 6000 photos of eight distinct fundus classes, using VGG-19, and ResNet50 image classification algorithms. Different ocular problems are characterized by these classifications. However, the dataset for these classes is somewhat erratic. This paper recommended using the same number of photos for both categories and creating a binary classification challenge out of this multiclass classification problem to address this difficulty. This paper also uses the latest vision transformer method. The models are trained with actual data as well as with images after applying LBP (local binary pattern) on the image.