An OGFA+CNN Approach for Multi-Level Disease Identification in Fundus Images

Fundus disease classification is critical for the early detection of conditions such as diabetic retinopathy and glaucoma, which can lead to blindness if left untreated. Timely and accurate diagnosis through fundus image analysis enables prompt intervention, personalized treatment, and better patient outcomes. Traditional methods often rely on manual feature extraction, which is labor-intensive and error-prone. While Convolutional Neural Networks (CNNs) have automated classification, they struggle to capture complex relationships in fundus images, especially subtle changes in blood vessels and the optic disc. Additionally, many CNN models fail to incorporate global context, which is essential for understanding the interdependencies between various image features. These limitations hinder CNN performance in accurately detecting early-stage abnormalities. The proposed Optimized Graph Filter Augmentation (OGFA) model addresses these issues by integrating custom graph filters with CNNs and other deep learning methods. Graph-based techniques are employed to capture the structural relationships between key elements such as blood vessels and the optic disc, providing valuable global context to the image. To enhance the model’s accuracy, custom-designed graph filters are used, improving the focus on important structural features and minimizing noise. This customization allows the OGFA model to go beyond local feature extraction capabilities of traditional CNNs and improves diagnostic precision. By incorporating graph-based methods for global context alongside CNNs for detailed local feature extraction, the OGFA model becomes more adept at detecting subtle abnormalities and intricate changes in fundus images. This hybrid approach boosts robustness and reliability, offering a more accurate and effective solution for fundus disease classification, particularly in clinical settings where precision is essential for optimal patient care.