Automated Glaucoma Detection Using Vision Transformers and Capsule Networks: A Deep Learning Approach for Early Diagnosis from Retinal Fundus Images

It is well known that glaucoma leads not only to visual impairment but also to total blindness, making its diagnosis very important in preserving the vision of affected patients. In this work, we offer a fully automated system employing Retinal fundus-based deep learning for glaucoma diagnosis where Vision Transformers and Capsule Networks are used in deep learning architectures. Suitable algorithms for upgrading the photo and enhancing the performance of the model are performed on the images, including plotting, histogram equalization, and normalization. The RIM-ONE dataset was achieved with a categorical cross-entropy loss and Adam Optimizer with 93% accuracy on the model. Grad-CAM is utilized to make the model explainable for the images and pinpoint the specific regions of images, which are responsible for the obtained predictions. The accuracy, precision, recall, and F1 score of the image are some of the performance evaluation criteria used. The aim of this approach is to devise a reliable and explainable solution for glaucoma primary diagnosis in a timely manner that would assist clinicians and advance the patient’s care.