IMPROVED SEGMENTATION AND CLASSIFICATION OF GLAUCOMA USING U-NET WITH DEEP LEARNING MODEL

Abstract

Glaucoma, a common eye illness that affects the optic nerve, is a major global health concern, with a
prevalence rate of 3.54 percent among those aged 40 to 80. Early identification is criticalin preventing
irreversible vision loss caused by optic nerve damage. The conventional techniques for early analysis
rely on analysing the optic cup and disc boundaries in fundus pictures, but they face obstacles such as
class imbalance, which reduces recognition quality. To address these challenges, a new solution
combines the U-Net architecture for fundus retinal image segmentation with Convolutional Neural
Networks (CNNs) for classification. This studyuses the U-Net architecture to segment glaucoma images,
taking use of its capacity to successfully collect detailed details and spatial information, improving the
precision of detecting optic cup and disc borders. Furthermore, the use of Convolutional Neural
Networks (CNNs) for classification allows for the differentiation of non-glaucoma and glaucoma patterns,
resulting in more accurate diagnostic conclusions. By combining both approaches, the study takes
advantage of the complimentary qualities of U-Net for segmentation and CNN for classification,
yielding a comprehensive and robust approach to glaucoma analysis. This integrated system attained an
impressive 90.8% accuracy rate, demonstrating its efficiency in enhancing the accuracy and reliability
of glaucoma diagnosis.