TL-GWO: Fine-tuned transfer learning with grey wolf optimizer for accurate fundus image-based eye disease...

Area of Science:

• Ophthalmology
• Computer Vision
• Artificial Intelligence

Background:

• Diabetic retinopathy and glaucoma are leading causes of vision loss.
• Early detection and diagnosis are crucial for effective treatment.
• Automated systems can aid in screening and diagnosis.

Purpose of the Study:

• To develop an automated diagnostic framework for detecting diabetic retinopathy, glaucoma, and healthy retinas.
• To optimize transfer learning (TL) models using the Grey Wolf Optimizer (GWO).
• To enhance image preprocessing and data augmentation for improved performance.

Main Methods:

• Utilized transfer learning backbones (DenseNet121, ResNet50, ResNet101V2, InceptionResNetV2, Xception).
• Employed Grey Wolf Optimizer (GWO) for architecture and hyperparameter optimization.
• Applied Contrast Limited Adaptive Histogram Equalization (CLAHE) and data augmentation.

Main Results:

• The ResNet101V2 model achieved the highest accuracy (89.32%) and F1-score (89.37%).
• GWO-driven optimization significantly improved model generalization and robustness.
• The framework outperformed other architectures across all evaluation metrics.

Conclusions:

• Combining TL strategies with metaheuristic optimization offers a reliable approach for ophthalmic disease detection.
• The proposed framework demonstrates potential for scalable computer-aided diagnostic systems.
• Automated detection systems can enhance the efficiency and accuracy of diagnosing eye diseases.