A Comparison of Deep Learning Techniques for Glaucoma Diagnosis on Retinal Fundus Images

Abstract

Glaucoma is one of the serious disorders which cause permanent vision loss if it left undetected. The primary cause of the disease is elevated intraocular pressure, impacting the optic nerve head (ONH) that originates from the optic disc. The variation in optic disc to optic cup ratio helps in early detection of the disease. Manual calculation of Cup to Disc Ratio (CDR) consumes more time and the prediction is also not accurate. Utilizing deep learning for the automatic detection of glaucoma facilitates precise and early identification, significantly enhancing the accuracy of glaucoma detection. The deep learning technique initiates the process by initially pre-processing the image to achieve data augmentation, followed by the segmentation of the optic disc and optic cup from the retinal fundus image. From the segmented Optic Disc (OD)and Optic Cup (OC) feature are selected and CDR calculated. Based on the CDR value the Glaucoma classification is performed. Various deep learning techniques like CNN, transfer learning, algorithm was proposed in early detection of glaucoma. From the comparative analysis glaucoma diagnosis, the proposed deep learning artifact Convolutional Neural Network outperform in early diagnosis of glaucoma providing accuracy of 99.3 8%.