Integrated Methodology for Early Glaucoma Diagnosis Using Retinal Fundus Images

This research explores a comprehensive methodology to glaucoma discovery using retinal fundus images, aiming to enhance diagnostic precision through advanced machine learning techniques. We begin with a diverse dataset of retinal images, applying bilinear interpolation for resizing, and convert images to greyscale using the Luminosity method to highlight relevant features. Noise is reduced using an adaptive Normalized Least Mean Squares (NLMS) filter, followed by histogram equalization to improve contrast. Feature extraction is accomplished using the method called Histogram of Oriented Gradients (HOG). These features are then classified with a combined model of ResNet-50 and Multiclass Support Vector Machine (MCSVM). The model’s efficacy is evaluated using precision, accuracy, recall, and F1-score metrics, demonstrating high effectiveness with an 98.66% of accuracy, 98.99% of precision, and 97.87% of recall. Future work will focus on expanding the dataset, integrating multi-modal data, and enhancing model explainability to further improve glaucoma detection accuracy.