Glaucoma detection using multi-feature extraction and classification techniques

Glaucoma is a progressive optic neuropathy and a major cause of irreversible blindness globally. In this work, fundus imaging is widely acknowledged as a vital technique for glaucoma diagnosis, as it provides detailed visualization of the retina. However, manual interpretation of these images is often time-consuming and subject to inter-observer variability. Conventional diagnostic approaches such as tonometry, perimetry, and optic nerve evaluation are resource-intensive and require expert interpretation, highlighting the demand for automated screening solutions. Thus, the proposed work introduces an integrated machine learning framework for glaucoma detection using fundus images. A curated dataset of 1,291 images from ACRIMA, RIM-ONE, and DRISHTI-GS was utilized. The pre-processing pipeline includes resizing, Gaussian filtering for noise removal, contrast enhancement via contrast limited adaptive histogram equalization (CLAHE), and synthetic minority over-sampling technique (SMOTE) for class balancing. Feature extraction combines handcrafted descriptors cup-to-disc ratio (CDR), histogram of oriented gradients (HOG), local binary patterns (LBP), and Gray-Level Co-occurrence Matrix (GLCM)—with deep features derived from pretrained convolutional neural networks such as ResNet-50, ResNet-101, and EfficientNet-B0. Classification is performed using Support Vector Machine (SVM), Random Forest, AdaBoost, and an Ensemble Voting Classifier. The framework compares models based on handcrafted features against those trained on deep pretrained features. The ensemble model using EfficientNet-B0 achieved the highest accuracy of 95%, while SVM and Random Forest achieved 94% and 92%, respectively, and AdaBoost reached 89%. In contrast, handcrafted feature-based models showed reduced accuracy. These results significantly surpass traditional techniques, which typically report 85–88% accuracy. The proposed work demonstrates that integrating multi-level feature extraction with ensemble learning significantly enhances classification performance and offers a promising, scalable solution for clinical glaucoma screening.