Diabetic Retinopathy and Glaucoma Diagnosis with CNN-Based Approach and User Interface for Morphological Analysis of Risky Regions in Fundus Images

Saim Ervural

doi:https://doi.org/10.54486/fivezero.2024.43

Saim Ervural, Emre Ergüler, Mehmet Akif Tokel

Year : 2025
Vol : 5
Issue : 1
Page : 22-44

DOI : https://doi.org/10.54486/fivezero.2024.43

Early diagnosis and effective monitoring of diabetic retinopathy and glaucoma are essential for preventing vision loss and improving patient outcomes. These conditions, if detected early, can be managed effectively, reducing the burden on healthcare systems and improving quality of life for patients. This study presents the development of a user-friendly software interface designed to assist healthcare professionals in diagnosing these conditions more efficiently. We developed a custom 11-layer Convolutional Neural Network (CNN) architecture, beginning with a rescaling layer and incorporating data augmentation techniques. The primary architecture consists of three convolutional layers containing 16, 32, and 64 filters, respectively, each followed by max pooling layers. A dropout layer with a 0.7 rate was incorporated to reduce the risk of overfitting. The network also features a flattening layer, a dense layer with 128 neurons for feature extraction, and an output layer tailored to the number of classes. For glaucoma detection, a specialized preprocessing step focusing on the optic disc reduced validation loss by approximately 20%. Additionally, a manual zooming feature was developed to enhance diagnostic accuracy in complex glaucoma cases. The algorithms for diabetic retinopathy were meticulously designed to identify and highlight pathological areas, such as edema and hemorrhage. This approach facilitates precise visualization of vascular structures and significantly enhances the model's capability to provide accurate and timely diagnoses. The architecture that emerged upon conclusion of the study demonstrated an accuracy of 98% for diabetic retinopathy and 85% for glaucoma. This study highlights the potential of advanced deep learning combined with practical tools to improve diagnostics, offering clinicians a reliable system to enhance patient outcomes.

Cite this Article As : Ervural, S., Ergüler, E., & Tokel, M. A. (2025). Diabetic Retinopathy and glaucoma diagnosis with CNN-based approach and user ınterface for morphological analysis of risky regions in fundus images. Fivezero, 5(1), 22-44. https://doi.org/10.54486/fivezero.2024.43

Description : Yazarların hiçbiri, bu makalede bahsedilen herhangi bir ürün, aygıt veya ilaç ile ilgili maddi çıkar ilişkisine sahip değildir. Araştırma, herhangi bir dış organizasyon tarafından desteklenmedi.Yazarlar çalışmanın birincil verilerine tam erişim izni vermek ve derginin talep ettiği takdirde verileri incelemesine izin vermeyi kabul etmektedirler. None of the authors, any product mentioned in this article, does not have a material interest in the device or drug. Research, not supported by any external organization. grant full access to the primary data and, if requested by the magazine they agree to allow the examination of data.

Diabetic Retinopathy and Glaucoma Diagnosis with CNN-Based Approach and User Interface for Morphological Analysis of Risky Regions in Fundus Images, Research Article,
2025, Vol. 5 (1)

Received : 13.11.2024, Accepted : 04.06.2025 , Published Online : 30.06.2025