AI-Driven Glaucoma Detection: A Deep Learning Approach with Explainable Diagnostics
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Glaucoma is one of the primary causes of irreversible blindness worldwide, emphasizing the importance of early and accurate diagnosis. Conventional glaucoma screening methods rely extensively on the expertise of ophthalmologists, making the process subjective, time-consuming, and prone to variability. To overcome these limitations, this study proposes an automated glaucoma detection system based on deep learning techniques for the analysis of retinal fundus images. Transfer learning is employed using state-of-the-art pretrained models, including ResNet50, VGG16, and EfficientNet. The performance of these models is optimized through comprehensive hyperparameter tuning, incorporating variations in learning rates, batch sizes, and the use of modern optimization algorithms such as AdamW. To enhance model explainability, Grad-CAM and Class Activation Mapping (CAM) are integrated, allowing visualization of critical image regions contributing to the prediction. Addition- ally, a GPT-based image model is incorporated to generate textual insights, improving the interpretability of results. Experimental evaluations compare the performance of different models and optimization techniques, ensuring a robust and clinically relevant AI-driven diagnostic system. Experimental results show that the proposed model achieved a higher classification accuracy and stability in different pretrained architectures. The findings highlight the potential of deep learning in automating Glaucoma screening, enhancing diagnostic accuracy and also shows strong potential for supporting scalable and transparent glaucoma screening in clinical settings.
Fields of Study & Subjects
Retinal Imaging and Analysis
Glaucoma and retinal disorders
Ophthalmology and Visual Health Research
Publication details
- Journal / source
- International Conference on Sentiment Analysis and Deep Learning (ICSADL)
- Pages
- 493–498
- Publication date
- 18 February 2026
- Language
- EN