A Robust Deep Learning-Based Classification Framework for Diabetic Eye Diseases Using Inception V3

Abstract

Diabetic eye diseases, including diabetic retinopathy (DR), diabetic macular oedema, cataracts, and glaucoma, are major causes of vision impairment worldwide. Early detection is critical, yet manual screening is time-consuming and prone to subjectivity. This paper proposes a robust deep learning-based classification framework using Inception V3 integrated with preprocessing, segmentation, and feature extraction techniques. Input retinal images undergo Adaptive Wiener Filtering for noise reduction, followed by Noise-Resilient Fuzzy C-Means (NR-FCM) segmentation. Local Binary Patterns (LBP) are applied for texture-based feature extraction, and histograms are generated to form feature vectors. These vectors are classified using the Inception V3 model, trained with parallel processing for efficiency. The proposed system demonstrates high accuracy and robustness, addressing challenges of class imbalance and interpretability. Results highlight improved diagnostic precision, supporting early detection and clinical decision-making.