Investigation of Glaucoma Prediction & Classification using Fundus Images with Machine Learning: A Comparative Study

Abstract

Glaucoma, a leading cause of irreversible blindness due to eye pressure, can be effectively managed with early prediction. Machine learning (ML) models use digital fundus images to extract intrinsic features for efficient glaucoma prediction at an early stage. This study mainly focus on investigation and comparison of the performance of machine learning algorithms such as Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Machine (GBM), and K-Nearest Neighbour (KNN) in predicting glaucoma and healthy classes using fundus images. Digital fundus images are utilized as an input for each method, with multiple features extracted to classify images into glaucoma and healthy categories using computational methods. The Drishti-DFI dataset is used for training and testing purposes. Performance metrics, including sensitivity, specificity, accuracy, and dice-coefficient, are employed to assess the performance of the prevailing ML models. The comprehensive results showcase different degrees of effectiveness across the machine learning models, with SVM exhibiting robust specificity, RF achieving balanced performance, GBM demonstrating superior accuracy, and KNN showing high sensitivity. The MATLAB tool is used to compare the models, which showcases the promising results and limitations of each model, offering insights into optimal strategies for glaucoma prediction using fundus images.