Pre-Trained Deep Learning-Based Approaches for Eye Disease Detection

This study primarily examines how well four pretrained deep learning models perform in identifying eye disorders using four metrics we developed: recall, precision, accuracy, and F1 Score. With the help of universal custom layers, the models are adjusted, and the outcomes are examined. The study then suggests an ensemble method that uses majority voting to combine the probabilistic outputs of the top-performing models. The suggested methodology outperforms state-of-the-art algorithms in experiments using a publicly available dataset, with average values for Recall, Precision, Accuracy, and F1 Score of 81.25%, 83.68%, 95.17%, and 79.12%, respectively. The work shows how well-trained deep learning models can identify eye illnesses and have the potential to improve public health, especially in mass screening programs.