Segmentation-Driven Glaucoma Classification from Retinal Fundus Images Using Optic Disk Localization, Linear and Nonlinear Multiple Instance Learning Frameworks

In recent years, glaucoma has been one of the leading causes of vision loss, and its early detection is essential for preventing irreparable damage. This study proposes an efficient classification approach using segmentation on retinal fundus images from the MESSIDOR dataset, focusing on optic disk segmentation to enhance disease-specific feature extraction while minimizing irrelevant background information. The pipeline includes preprocessing steps such as image resizing and green channel extraction, followed by segmentation of both the optic cup and optic disk using traditional image processing techniques, including pixel clustering with superpixel methods, global thresholding, and the Circular Hough Transform (CHT). Performance is evaluated using metrics such as testing correctness (TC), sensitivity, specificity, F1-score, and computational time under 5-fold, 10-fold, and Leave-One-Out (LOO) cross-validation (CV). Results show that preprocessing significantly improves accuracy across all models, with mi-SPSVM under 10-fold CV achieving the best results, obtaining 90.00% sensitivity and a 67.09% F1-score. Meanwhile, mi-KSPSVM achieved the lowest computational time, processing a single image in just 0.09s. These findings demonstrate that incorporating segmentation into the classification task can enhance both diagnostic accuracy and computational efficiency, making the approach suitable for real-time applications and resource-limited environments. Overall, this study presents a promising and scalable solution for automated OD screening, with future work aimed at integrating deep learning-based segmentation to improve precision further.