Applied Explainable Artificial Intelligence (XAI) in the classification of retinal images for support in the diagnosis of Glaucoma

Machine learning models have become ubiquitous across various domains, revolutionizing disease diagnosis and offering remarkable applications in healthcare. In particular, the use of artificial intelligence techniques has significantly transformed the field of ophthalmology, aiding in the early detection of neurodegenerative eye diseases like glaucoma through image classification. However, the lack of explainability in model decisions poses a substantial barrier to their widespread adoption in clinical practice. This research addresses this limitation by exploring and applying explainable artificial intelligence (XAI) techniques to different convolutional neural network (CNN) architectures for glaucoma classification. Our study focuses on providing ophthalmologists with robust resources for human interpretation and supporting clinical diagnosis. We propose an innovative visual interpretation approach called SCIM (SHAP-CAM Interpretable Mapping) and compare its performance against existing techniques, such as Gradient-Weighted Class Activation Mapping (Grad-CAM). Our experiments, conducted on the VGG19 architecture, demonstrate that both Grad-CAM and the novel SCIM approach offer superior resources for human interpretability, further enhancing the potential of CNNs in glaucoma diagnosis.