Fundus Image-Based Eye Disease Detection Using EfficientNetB3 Architecture

Area of Science:

• Ophthalmology
• Computer Science
• Artificial Intelligence

Background:

• Early detection of retinal diseases is crucial for preventing vision loss.
• Manual diagnosis of fundus images is time-consuming and prone to errors.
• Automated solutions are needed to improve diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate a deep learning model for automated classification of retinal diseases.
• To categorize fundus images into four classes: cataract, diabetic retinopathy, glaucoma, and healthy.
• To assess the model's performance using various classification metrics.

Main Methods:

• Utilized a pretrained EfficientNetB3 architecture for image classification.
• Fine-tuned the model on a public Kaggle retinal image dataset.
• Employed transfer learning, data augmentation, and the Adam optimizer with a cosine annealing scheduler.

Main Results:

• Achieved a high classification accuracy of 95.12%.
• Demonstrated strong performance with precision (95.21%), recall (94.88%), F1-score (95.00%), Dice Score (94.91%), Jaccard Index (91.2%), and MCC (0.925).
• The model showed robustness in classifying four distinct retinal conditions.

Conclusions:

• The proposed deep learning model shows significant potential for automated retinal disease diagnosis.
• This automated system can support clinical decision-making and improve patient outcomes.
• Further validation in clinical settings is warranted to confirm its utility.