Development of an Enhanced Retinal Disease Predictive System using Convolutional Neural Network
Article PDF

Keywords

Central serous retinopathy
convolutional neural networks
models
prediction system
retinal disorder

How to Cite

Oboro, E., Akazue, M., Abel, E., & Akpojaro, J. (2025). Development of an Enhanced Retinal Disease Predictive System using Convolutional Neural Network. Global Journal of Artificial Intelligence and Technology Development , 3(3), 17-30. https://www.openjournals.ijaar.org/index.php/gjaitd/article/view/1309

Abstract

Retinal diseases such as diabetic retinopathy, glaucoma, central serous retinopathy, age-related macular degeneration (AMD), and retinitis remained among the primary causes of visual impairment and blindness worldwide. Early and accurate detection plays a vital role in preventing irreversible vision loss and improving patient outcomes. Despite the  significant advancements with the integration of machine learning techniques,  automated retinal disease diagnosis still rely heavily on traditional machine learning models such as Support Vector Machines (SVM) for classification tasks, and .predicted one retinal disease. Thus, such SVM models suffer low-dimensional data, and their performance in handling complex high-dimensional retinal images is limited. Therefore, this research presents the development of an enhanced retinal disease predictive system using Convolutional Neural Networks (CNNs) to classify retinal fundus images into four categories three disease types and a normal class. The system was designed and implemented using Python programming language, utilizing libraries such as TensorFlow, Keras, NumPy, Matplotlib, and OpenCV. The development process followed structured phases: Requirement Analysis, System Design, Implementation, Testing, Deployment, and Maintenance. The system also incorporated a mathematical model of CNN layers including convolution, activation (ReLU), pooling, and fully connected layers optimized to learn critical visual patterns from retinal images. A comprehensive preprocessing pipeline was further implemented to enhance image clarity, normalize lighting variations, and standardize input size and format. The model was trained on a dataset of 4,217 images (sourced from Kaggle), split into training, validation, and test sets. The evaluation of this CNN-based model achieved a classification accuracy of 91%, with precision, recall, and F1-scores consistently above 88% across all classes. The system supports both static image uploads and real-time webcam capture, enabling flexible use in clinical or field settings. Evaluation through confusion matrix analysis revealed minimal false predictions, demonstrating high reliability. This study confirms the potential of CNN-based systems in automating retinal disease diagnosis and supporting ophthalmologists with faster, more accurate screening tools.

Article PDF
Creative Commons License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

Copyright (c) 2025 Global Journal of Artificial Intelligence and Technology Development

376 191