Feature Extraction And Classification Of Glaucoma Using Retinal Images

Abstract

Glaucoma, the leading cause of permanent blindness worldwide, must be diagnosed as soon as possible to avoid further visual loss. Retinal imaging, particularly of the optic disc and cup, can assist diagnose glaucoma. To detect glaucoma, these factors must be carefully separated. This paper describes a unique approach to glaucoma detection based on retinal image feature extraction and classification. Following retinal image segmentation, the proposed method extracts features using RESNET and ViT to separate the optic disc and cup. RESNET uses residual learning to acquire deep features in order to learn complicated patterns, but ViT increases feature extraction by emphasizing global contextual information using self-attention approaches. These additional strategies allow one to extract both local and global elements from the various regions of interest. After getting the features, the data is classified using an upgraded VGG19 model. VGG19, a prominent convolutional neural network design, has been tuned for glaucoma case classification. The update's network tuning helps the system discriminate between glaucomatous and healthy alterations in retinal images. The obtained features can be used with the improved VGG19 model to identify images as glaucomatous or non-glaucomatous. Experimental results on publicly available retinal imaging datasets demonstrate higher classification accuracy, sensitivity, and specificity than cutting-edge approaches. By combining RESNET with ViT for feature extraction and updated VGG19 for classification, the glaucoma detection system becomes more stable and efficient. This approach has the potential to transform early diagnosis and reduce the risk of blindness associated with fast therapy through routine glaucoma screening.