Development of an Ocular Disease Prediction Using Deep Learning

• Toyin OKEBULE

  Author

Ocular diseases remain a leading cause of visual disability, significantly affecting quality of life and productivity, particularly in low- and middle-income countries with limited access to specialized eye care. Early and accurate diagnosis relies heavily on experienced ophthalmologists, despite advancements in imaging techniques such as fundus photography, optical coherence tomography (OCT), and ultrasound. This study presents a hybrid deep learning framework for multi-class ocular disease classification, leveraging Convolutional Neural Networks (CNNs) for spatial feature extraction and ensemble methods for classification refinement. The dataset employed consists of 7,230 ocular images across five disease categories. The dataset includes patients aged 30–80 years, with 55% male and 45% female, predominantly of Nigerian African ethnicity. All images are fundus photographs (no OCT or multimodal images), ensuring consistent imaging modality. The dataset consists of diabetic retinopathy (1,750 images), glaucoma (1,420), macular degeneration (1,260), cataracts (1,100), and normal images (1,700). Extensive preprocessing, including resizing, normalization, contrast enhancement and data augmentation was a pplied to enhance model robustness and generalization. The model was trained and validated using stratified splits (70% training, 15% validation, 15% testing). The ensemble approach outperformed individual models, achieving 97.8% accuracy, with a classification report confirming minimized false positives and false negatives, critical in clinical diagnostics. The model also demonstrated low inference time and high computational efficiency, supporting potential deployment in clinical decision-support systems and mobile diagnostic applications.

• 5_Okebule (2026)

Copyright (c) 2026 Toyin OKEBULE (Author)

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