International Journal of Computational and Electronic Aspects in Engineering Volume 6, Issue 3, July 2025, pp 168 - 179 https://doi.org/10.26706/ijceae.6.3.20250607 168 Int. J. of Computational and Electronic Aspects in Engineering https://www.rame.org.in/ijceae/ Prospective Detection of Diabetic Retinopathy Using Modified CNN Models on Fundus Images: A Study at Al-Noor Institute, Al-Nasiriya Atyaf Jarullah Yaseen Department of Computer Science, College of Computer Science and Mathematics, University of Thi-Qar, Iraq Correspondence: atyafjarallah82@utq.edu.iq Abstract: Diabetic retinopathy (DR) is considered to be the most common microvascular complication with diabetes mellitus and continues to be the main disease that causes vision impairment and blindness all over the world. Early-stage outcomes are, however, difficult to identify; they require highly qualified clinical retinal fundus photos interpretation to detect DR at an opportune stage in order to avert visual disability that would most probably be irreversible. The aim of this prospective study was to consider a deep learning-based diagnostic model, which is based on the modified convolutional neural network (CNN) trained and tested on a proprietary dataset and estimated in Al-Noor Institute in Al-Nasiriya in the Ophthalmology department. The goal of the model was to evaluate the quality of input fundus images, and group them in such categories as DR-positive, and DR-negative. Clinical ophthalmologists were used to check the production of the model and certify the results of model accuracy. The study used 398 patients (232 males and 166 femals) screened over a five weeks period. Compared to the expert-labeled ground truth, the proposed model had an accuracy of 93.72%, sensitivity of 97.30%, and a specificity of 92.90% initialization. This evidence underlines the feasibility of deep learning applications in helping to detect diabetic retinopathy early and especially in low-resource environments. Keywords: Diabetic retinopathy. Fundus imaging, convolutional neural networks, deep learning, automated diagnosis, ophthalmology. 1. Introduction Diabetic retinopathy (DR) has been identified as the most common microvascular complication attributed to diabetes mellitus and one of the major causes of blindness and sight deficiency in different parts of the world. Epidemiologic estimates indicate that more than 200 million will be afflicted by this condition by the year 2040 [1]. The destruction of the small blood vessels within the retina, or light-sensitive tissue at the back of the eye, is the main work of DR and causes gradual loss of vision. Studies have shown DR to fall in two clinical categories of: Non-Proliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy (PDR). The initial phases of the condition (NIH NPDR), retinal microvasculature is destroyed as a result of swelling and rupture of the capillaries. This condition commonly leads to macular edema, where the central part of the retina (macula) is swollen and this is one of the reasons of mild to moderate visual loss. By contrast, PDR manifests the advanced form of the disease, during which unusual neovascularization takes place on the retinal surface. These weak new vessels tend to burst, resulting in hemorrhage to the vitreous, fibrosis and in the end serious loss of vision. Some risk factors are related to the emergence and development of DR among which it is possible to distinguish type 1 and type 2 diabetes, hypertension, hyperlipidemia, pregnancy, the ethnicity, and positive family history. Article – Peer Reviewed Received: 1 June 2025 Accepted: 11 July 2025 Published: 25 July 2025 Copyright: © 2025 RAME Publishers This is an open access article under the CC BY 4.0 International License. https://creativecommons.org/licenses /by/4.0/ Cite this article: Atyaf Jarullah Yaseen, “Prospective Detection of Diabetic Retinopathy Using Modified CNN Models on Fundus Images: A Study at Al-Noor Institute, Al-Nasiriya”, International Journal of Computational and Electronic Aspects in Engineering, RAME Publishers, vol. 6, issue 3, pp. 169-179, 2025. https://doi.org/10.26706/ijceae.6.3.2 0250607