Multi-Dataset Convolutional Neural Network Model for Glaucoma Prediction in OCT Fundus Scans

Abstract

Glaucoma, a major factor in permanent blindness across the globe, necessitates accurate and efficient diagnostic methods. This article presents a comprehensive approach to glaucoma prediction by combining three diverse datasets: ORIGA, ACRIMA, and REFUGE. A novel multi-Datasets CNN (MD-CNN) architecture is proposed, specifically tailored to effectively handle combined data comprising diverse image characteristics across multiple datasets. This innovative approach demonstrates remarkable robustness in accommodating variations in image attributes, including lighting, zoom levels, and other disparate features, thus showcasing its potency in addressing glaucoma prediction across different datasets. The approach demonstrates improved accuracy (96.88%), sensitivity (94.34%), specificity (97.20%), precision (94.34%), and AUC (99.02%) compared to individual dataset-based models, addressing challenges in glaucoma detection. This research showcases the potential of combining diverse datasets for more effective CNN-based glaucoma detection.