Automated Tool Support for Glaucoma Identification with Explainability Using Fundus Images

The proposed study presents an automated tool for glaucoma identification utilizing fundus images, emphasizing the importance of explainability in machine learning models. The process begins with image pre-processing, employing two primary techniques: Contrast Limited Adaptive Histogram Equalization (CLAHE) and median filtering, to enhance image quality and remove noise. Data augmentation techniques, including rotation and vertical flipping, are applied to increase the robustness of the model by diversifying the training dataset. Subsequently, the U-Net architecture is utilized for precise segmentation of the optic disc and cup in fundus images, enabling the model to focus on relevant features indicative of glaucoma. The classification of healthy and glaucomatous eyes is performed using the InceptionV3 deep learning model, known for its superior performance in image recognition tasks. To enhance the interpretability of the model, Grad-CAM (Gradient-weighted Class Activation Mapping) is implemented, providing visual explanations of the model's predictions by highlighting important regions in the fundus images. The overall accuracy of the automated tool is evaluated, demonstrating its potential as a reliable support system for ophthalmologists in early glaucoma detection, ultimately contributing to improved patient outcomes through timely diagnosis and intervention.

Index Terms— Artificial intelligence, classification, explainability, segmentation, support tool, trustworthiness.