Abstract

Diabetic retinopathy (DR) is one of the leading causes of vision loss globally, particularly among diabetic patients. Early and accurate grading of DR is critical for timely intervention and effective management of the disease. However, the variability in retinal lesion patterns and the high-dimensional nature of retinal image data present significant challenges in achieving precise classification. To address these challenges, a multistage framework integrating a Hybrid Fuzzy-KNN (HF-KNN) classifier is proposed for DR grading. The framework begins with preprocessing techniques to enhance retinal image quality by reducing noise and enhancing contrast. Following this, region-specific feature extraction techniques are applied to capture clinically relevant features such as microaneurysms, exudates, and hemorrhages. The extracted features are normalized and reduced in dimensionality using Principal Component Analysis (PCA) to optimize computational efficiency. The proposed Hybrid Fuzzy-KNN classifier employs fuzzy logic to handle uncertainty in feature classification and combines it with the simplicity of KNN for efficient grading into five stages of DR: no DR, mild, moderate, severe, and proliferative DR. A benchmark dataset of retinal images is utilized for evaluation, achieving an overall classification accuracy of 96.7%, sensitivity of 94.8%, specificity of 97.5%, and F1-score of 95.2%. The system outperforms traditional KNN and fuzzy-based methods, demonstrating its robustness in handling complex retinal data.

Authors

Nagamany Abirami, K. Andal
Sri Venkateshwara College of Engineering and Technology, India

Keywords

Diabetic Retinopathy, Hybrid Fuzzy-KNN, Multistage Framework, Retinal Image Analysis, Classification Accuracy