|
|
 |
| |
|
|
|
A Deep Learning-Based Eye Disease Diagnosis Using OCT Imaging and SE-Enhanced CNNs |
|
|
|
PP: 1153-1165 |
|
|
|
doi:10.18576/amis/190515
|
|
|
|
Author(s) |
|
|
|
Amani A. Slamaa,
|
|
|
|
Abstract |
|
|
| Vision loss remains a major global health concern, with cataract, glaucoma, diabetic retinopathy (DR) being leading, yet often symptomless, causes of preventable blindness, highlighting the urgent need for early, accessible, and cost-effective diagnostic solutions. This study introduces DEEPSIGHT, a deep learning-powered diagnostic system designed to automatically detect these three diseases using optical coherence tomography (OCT) imaging. The system aims to deliver both high diagnostic accuracy and practical usability in clinical settings. At its core is custom convolutional neural network (CNN) architecture, enhanced with attention mechanisms such as squeeze-and-excitation (SE) blocks to improve feature extraction. The model was trained on a diverse dataset of OCT images collected from public sources and clinical partners. Preprocessing steps—including normalization, contrast enhancement and data augmentation—were applied to improve robustness and reduce over fitting. A stratified 5-fold cross-validation strategy was used during training, with categorical cross-entropy loss and the Adam optimizer. DEEPSIGHT achieved over 94% accuracy, with precision and F1-scores exceeding 92% across all classes. To support clinical interpretability, Grad-CAM and saliency maps were integrated, allowing visualization of the image regions influencing model predictions. The system was deployed in a prototype diagnostic platform and validated on an independent clinical dataset, confirming its reliability and real-world applicability. While currently limited to three diseases and local deployment, future work will focus on cloud integration, broader diagnostic coverage, and real-time teleophthalmology support to enhance accessibility and scalability. This research contributes to the growing field of AI in healthcare and underscores the transformative potential of deep learning in vision science. |
|
|
|
|
|
|
|
