Automatic Glaucoma Classification and Justification Using a Large and Diverse Dataset

Glaucoma, a group of eye diseases causing optic nerve damage, can lead to irreversible vision loss if undetected. Fundus imaging is crucial for glaucoma diagnosis, but manual analysis is costly and inconsistent. This paper presents a method leveraging the extensive JustRAIGS challenge dataset to train two Convolutional Neural Networks (CNNs) for automatic glaucoma detection and feature identification. The referral model identifies images with signs of referable glaucoma, while the justification model detects ten specific glaucomatous features. As a preprocessing step, a pre-trained segmentation model isolates the optic disc region for focused analysis. By addressing class imbalance through sampling techniques and employing pre-trained CNN architectures, our method demonstrates superior performance in both tasks when compared to existing models. Gradient activation maps reveal enhanced feature capture around the optic disc, validating our approach's effectiveness in real-world, heterogeneous datasets. Our results are available on GitHub: https://github.com/kisandre7/sipaim2024-giaucoma