Segmentation of optic disk for glaucoma using attention U-Net deep learning model

This thesis addresses optic disk segmentation using deep learning techniques to support early diagnosis of glaucoma and accelerate the diagnostic process. Glaucoma is a common eye disease that can lead to vision loss and can cause permanent damage if not diagnosed early. The optic disk is a critical structure in glaucoma diagnosis as it contains the exit point of the optic nerve fibers. In this study, a customized attention UNet model is employed for optic disk segmentation, based on the UNet architecture. In the encoder part, in addition to the traditional UNet, Channel Attention and Patch-Based Attention mechanisms are integrated. Spatial Attention is used in the decoder part. This allows the model to focus more accurately on important features within the image, aiming to improve segmentation performance. These customized attention mechanisms enable the model to emphasize more prominent features and achieve more accurate results. The proposed model aims to provide accurate and fast segmentation to assist in glaucoma diagnosis by combining deep learning and image processing techniques. The results of the relevant study demonstrate that the proposed model achieves high accuracy and sensitivity. Compared to existing methods, the model's ability to improve the glaucoma diagnosis process and its potential for clinical applications are highlighted. This thesis represents an important step towards early diagnosis and treatment of glaucoma, aiming to prevent vision loss. Future studies could focus on expanding the model and testing it with different data sources to increase its applicability across a wider range of patients.