Enhancing Glaucoma Diagnosis with Explainable AI Using Vision Transformers and Advanced Deep Learning Techniques

Glaucoma is a leading cause of irreversible blindness worldwide, necessitating early and accurate diagnosis to prevent vision loss. To ensure trustworthiness in the accuracy of diagnosis, interpretability, and understanding of risk factors associated with glaucoma are essential. This study proposes a comprehensive methodology for glaucoma classification using various deep learning architectures, including CNN, VGG16, VGG19, InceptionResNetV2, Xception, and Vision Transformers (ViTs) integrated with Explainability techniques. The dataset utilized for this research comprises retinal fundus images divided into training, validation, and testing sets. The approach involves applying various preprocessing techniques, like CLAHE, green channel conversion, and canny edge detection, to enhance feature extraction. Among the evaluated models, the ViT demonstrated superior performance, achieving an accuracy of 92% on the test set. This model’s ability to dynamically adjust attention weights based on the input image contributes to its improved classification capabilities. To provide insights into the model’s decision-making process, Grad-Cam facilitates better treatment planning for clinicians.