Advancing Glaucoma Detection Using Deep Learning Techniques

Abstract

Glaucoma is a progressive optic neuropathy and one of the leading causes of irreversible blindness worldwide due to its asymptomatic nature during early disease stages. Early detection and timely intervention are essential for preventing permanent visual impairment; however, conventional diagnostic procedures rely heavily on expert interpretation, specialized clinical equipment, and time-intensive examination methods, thereby limiting large-scale screening. This research presents a deep learning-based automated glaucoma detection framework developed using retinal fundus images with a focus on multi-class disease classification. The proposed system employs a convolutional neural network architecture capable of automatically extracting hierarchical visual features associated with glaucomatous damage without the need for handcrafted feature engineering. A structured methodological pipeline involving dataset preprocessing, normalization, data augmentation, model training, validation, and comprehensive performance evaluation was implemented to ensure reliability and robustness. Experimental results demonstrate that the proposed model achieves an overall classification accuracy of 78.73%, with macro precision of 0.7889, macro recall of 0.7873, and macro F1-score of 0.7866, indicating balanced and unbiased predictive behaviour. Confusion matrix analysis reveals strong recognition of visually distinctive glaucoma categories, while misclassification occurs primarily in early or borderline disease stages due to inherent clinical ambiguity. Training and validation learning curves confirm stable convergence and effective generalization, highlighting the potential of deep learning-assisted glaucoma screening systems as scalable clinical decision-support tools.

Introduction

Glaucoma is a progressive eye disease that damages the optic nerve and often remains symptomless until significant vision loss occurs. With an aging global population, its prevalence is expected to rise, making early detection increasingly important. However, conventional diagnostic methods (e.g., intraocular pressure tests, visual field analysis, and OCT imaging) require specialized equipment and expertise, limiting their use in large-scale or resource-poor screening.

To address these limitations, the study focuses on automated glaucoma detection using retinal fundus images and deep learning, particularly convolutional neural networks (CNNs). These models can learn directly from raw images and identify complex patterns linked to glaucoma, reducing reliance on manual feature extraction. The system also supports multi-class classification, enabling detection of different disease stages rather than only normal vs. glaucoma, which improves clinical usefulness for monitoring progression and treatment planning.

The literature review shows a shift from traditional image processing and machine learning methods toward deep learning approaches, which generally provide higher accuracy and better feature representation. Key improvements include transfer learning, data augmentation, and class imbalance handling. Despite this progress, challenges remain in detecting early-stage glaucoma, ensuring model interpretability, managing dataset bias, and achieving reliable real-world deployment across diverse populations.

The proposed methodology uses a balanced dataset of retinal images, standardized preprocessing, and a CNN-based architecture for feature extraction and classification. The system pipeline includes image acquisition, preprocessing (normalization, resizing, augmentation), CNN feature learning, and softmax-based multi-class output generation. Performance is evaluated using accuracy, precision, recall, F1-score, confusion matrix, and learning curves to ensure comprehensive assessment.

Experimentally, the model achieved about 78.73% accuracy across ten classes, with balanced precision and recall (~0.79). It performed well on clear, advanced-stage cases but struggled more with early and intermediate stages due to subtle and overlapping retinal features. Confusion matrix analysis showed most errors occurred between adjacent disease stages, which reflects real clinical ambiguity rather than random failure.

Conclusion

This research presented a comprehensive deep learning–based framework for automated multi-class glaucoma detection using retinal fundus image analysis, addressing several critical limitations associated with traditional diagnostic approaches. Glaucoma remains one of the leading causes of irreversible blindness worldwide, largely due to delayed detection resulting from its asymptomatic progression during early disease stages. Conventional diagnostic methods, while clinically effective, are often dependent on specialist expertise, subjective interpretation, and costly imaging infrastructure, thereby restricting large-scale screening and timely intervention. In this context, the proposed deep learning framework offers a promising solution by enabling objective, scalable, and efficient glaucoma classification capable of supporting modern ophthalmic screening initiatives. The experimental findings demonstrate that the developed convolutional neural network model achieves balanced predictive performance across multiple diagnostic classes, reflecting its capability to learn meaningful hierarchical representations of glaucomatous structural patterns. The achieved overall accuracy of 78.73 percent, accompanied by closely aligned macro precision, recall, and F1-score values, indicates consistent classification behaviour without strong bias toward specific categories. Such balanced performance is particularly significant in medical screening applications, where both false negatives and false positives carry substantial clinical implications. The confusion matrix analysis further revealed clinically realistic misclassification patterns, primarily occurring between visually adjacent or early-stage glaucoma categories. These results highlight the inherent diagnostic complexity of glaucoma staging and reinforce the importance of automated systems functioning as decision-support tools rather than replacements for expert clinical judgment. Another important contribution of this study lies in the analysis of training stability and generalization capability. The convergence patterns observed in training and validation accuracy and loss curves confirm that the proposed architecture effectively balances learning capacity and regularization, thereby reducing the risk of overfitting. Stable learning behaviour enhances confidence in the model’s ability to perform reliably on previously unseen retinal images, which is essential for practical deployment in real-world healthcare environments. Furthermore, the multi-class formulation adopted in this research improves clinical relevance by enabling differentiation among multiple disease stages, thereby supporting treatment planning and longitudinal patient monitoring. Despite these promising outcomes, the study acknowledges certain limitations that provide direction for future research. Improving sensitivity in early-stage glaucoma detection remains a key priority, as subtle structural variations continue to pose challenges even for advanced deep learning models. Future investigations may explore the integration of multimodal diagnostic data, such as optical coherence tomography measurements, visual field assessments, and intraocular pressure readings, to enhance predictive accuracy and clinical robustness. Additionally, the incorporation of explainable artificial intelligence techniques may improve interpretability and facilitate clinician trust in automated diagnostic outputs. Validation across diverse datasets, imaging devices, and demographic populations will also be essential to ensure equitable performance and generalizability. Overall, this research contributes to the advancement of intelligent ophthalmic diagnostic systems by demonstrating the practical feasibility of deep learning–assisted glaucoma screening. By enabling timely detection, supporting clinician decision-making, and enhancing screening scalability, the proposed framework has the potential to play a meaningful role in reducing preventable vision loss and improving global eye health outcomes.

References

[1] Chen, X., Xu, Y., Wong, D. W. K., Wong, T. Y., & Liu, J. (2015). Glaucoma detection based on deep convolutional neural network. IEEE Transactions on Biomedical Engineering, 62(7), 1866–1876. [2] Hemelings, R., Elen, B., Barbosa-Breda, J., Stalmans, I., & Van Ginneken, B. (2021). Deep learning on fundus images detects glaucoma beyond the optic disc. Scientific Reports, 11(1), 17228. [3] Ajitha, S., & Balamurugan, S. (2021). Automatic glaucoma diagnosis using deep convolutional neural network. Journal of Ambient Intelligence and Humanized Computing, 12(6), 6335–6344. [4] Saha, S., et al. (2023). A fast and fully automated CNN-based system for glaucoma detection using fundus images. Scientific Reports, 13, 16345. [5] Yousefi, S., Goldbaum, M. H., & Balasubramanian, M. (2023). Clinical applications of artificial intelligence in glaucoma. Frontiers in Medicine, 10, 112233. [6] Ashtari-Majlan, M., et al. (2024). Glaucoma diagnosis in the era of deep learning: A comprehensive survey. Expert Systems with Applications, 245, 122771. [7] Chiang, Y. Y., et al. (2024). Deep learning evaluation of glaucoma detection using fundus photographs. Biomedicines, 12(7), 1394. [8] Meedeniya, D., et al. (2025). Deep learning approaches for glaucoma identification using retinal fundus images: A systematic review. Array, 20, 100311. [9] Ling, X. C., et al. (2025). Deep learning in glaucoma detection and progression prediction: A meta-analysis. Biomedicines, 13(2), 420. [10] Vuppu, V. M., et al. (2024). Early glaucoma detection using hybrid CNN-LSTM framework. Engineering, Technology & Applied Science Research, 14(3), 13422–13429. [11] Roy, A., Sen, A., Gupta, S., Haldar, S., & Das, A. (2025). DeepEyeNet: Hybrid deep learning framework for automated glaucoma detection. arXiv preprint arXiv:2501.11168. [12] Abramovich, O., et al. (2025). GONet: A generalizable deep learning model for glaucoma optic neuropathy detection. arXiv preprint arXiv:2502.19514. [13] García, G., del Amor, R., Colomer, A., Verdú-Monedero, R., & Naranjo, V. (2021). Hybrid prototypical neural networks for glaucoma grading using OCT. IEEE Access, 9, 78956–78968. [14] Maetschke, S., Antony, B., Ishikawa, H., Wollstein, G., Schuman, J. S., & Garnavi, R. (2019). Feature-agnostic deep learning for glaucoma detection in OCT volumes. IEEE Transactions on Medical Imaging, 38(2), 421–431. [15] Thiery, A. H., Braeu, F., Tun, T. A., Aung, T., & Girard, M. J. A. (2022). Geometric deep learning for glaucoma diagnosis. Medical Image Analysis, 75, 102265. [16] Gulshan, V., Peng, L., Coram, M., et al. (2016). Development and validation of a deep learning algorithm for detection of diabetic retinopathy. JAMA, 316(22), 2402–2410. [17] Ting, D. S. W., Cheung, C. Y., Lim, G., et al. (2017). Development and validation of deep learning system for diabetic retinopathy and related eye diseases. JAMA, 318(22), 2211–2223. [18] Orlando, J. I., Prokofyeva, E., & Del Fresno, M. (2020). Glaucoma classification using deep features. Pattern Recognition Letters, 130, 73–80. [19] Li, Z., He, Y., Keel, S., et al. (2018). Efficacy of deep learning in detecting glaucomatous optic neuropathy. Ophthalmology, 125(8), 1199–1206. [20] Raghavendra, U., Bhandary, S. V., Gudigar, A., et al. (2018). Automated diagnosis of glaucoma using deep learning. Computers in Biology and Medicine, 93, 266–274. [21] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. [22] Krizhevsky, A., Sutskever, I., & Hinton, G. (2012). ImageNet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 25. [23] Simonyan, K., & Zisserman, A. (2015). Very deep convolutional networks for large-scale image recognition. International Conference on Learning Representations. [24] Szegedy, C., Liu, W., Jia, Y., et al. (2015). Going deeper with convolutions. IEEE Conference on Computer Vision and Pattern Recognition, 1–9. [25] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. IEEE Conference on Computer Vision and Pattern Recognition, 770–778.

Copyright

Copyright © 2026 Priya Tiwari, Seema Kirar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.