Detection of Glaucoma Based on Retinal-Image Analysis Using Convolution Neural Networks Technique

Glaucoma is perennially progressive optic nerve disorder, which leads to permanent loss of vision. But vision loss from glaucoma can often be prevented with early treatment. Traditional diagnostic techniques concentrate on lowering intraocular pressure. We offer a deep learning-based technique for detecting visual impairment using fundus pictures. Specifically, the optic disc and cup is focused on during the segmentation phase using an attention-augmented U-Net design. Furthermore, ResNet50 is incorporated as a feature extractor into the U-Net architecture, enhancing segmentation accuracy and enabling a precise image of the optic cup and disc boundaries. An InceptionV3architecture that has been altered to differentiate between glaucomatous and healthy eyes is being used for classification. With its explainability to extract multiscale characteristics from fundus images, our deepCNN model can distinguish between normal retinal states and the delicate patterns with the stages that indicate glaucoma with high accuracy. To make sure the model generates precise and understandable prediction, we also employ interpretation techniques like the Grad-CAM method and Grad-CAM technology++.