Glaucoma Detection and Classification using Innovative Approaches based on Deep Learning model

Abstract

This study rigorously evaluates the efficacy of several deep learning models across multiple datasets, including RIM-ONE v3, ORIGA-Lite, and APTOS 2019, each known for their pivotal role in ophthalmic research. The models under examination—AlexNet, VGGNet 19, VGGNet 22, ResNet 101, and ResNet 152—were assessed on their performance across a comprehensive set of metrics: Accuracy, Sensitivity, Specificity, Precision, F1-score, and AUC. Our findings reveal a clear hierarchy in model performance, with ResNet 152 consistently outperforming its counterparts across all datasets, achieving remarkable scores such as a 98% accuracy, 97% sensitivity, and 98% specificity on the RIM-ONE v3 dataset, and similarly high metrics on the ORIGA-Lite and APTOS 2019 datasets. These results not only underscore the superiority of ResNet 152 in mastering the complex task of glaucoma detection but also highlight the model's adeptness at generalizing across different imaging conditions and patient demographics. The study's outcome suggests a promising direction for integrating advanced convolutional neural network architectures into clinical settings, offering a potential breakthrough in early glaucoma diagnosis and contributing significantly to the prevention of irreversible vision loss through timely and accurate classification.