Presumptive discerning of the severity level of glaucoma through clinical fundus images using hybrid PolyNet. (March 2023)

• Record Type: Journal Article
• Authors: Sangeethaa, S.N.
• Is Part Of: Biomedical signal processing and control. Volume 81(2023)
• Publication Date: 2023-03
• Subjects: Blood vessel segmentation -- Glaucoma -- Hybrid Polynet -- Optic disc -- Optic cup; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104347
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Purpose: Glaucoma is the major leading cause of non-curable visual defects and blindness across the world. Glaucoma is one of the eye diseases which cause an optic nerve, due to increased pressure called intraocular pressure. If the damage gets worse, it causes permanent vision loss within a few years. Timely diagnosis, early and accurate prediction are the most important factors for the prevention of visual blindness. For experts, it is uncompromising task for detecting glaucoma manually. In most of the retinal images, different investigations are carried out for identifying an automated detection tool for diagnosing glaucoma. Methods: The main aim of this paper is to build a machine learning model for predicting glaucoma and its severity level. To predict the severity level of glaucoma, firstly, the segmentation of the blood vessel is performed and classification is accomplished using proposed U-Net model. Secondly, the segmentation of the Optic Disc (OD) and the Optic Cup (OC) is performed to determine the Cup to Disk Ratio (CDR). The proposed hybrid PolyNet classifies the severity of the glaucoma disease. The earlier detection of glaucoma disease is important for preventing the severity of the eye disease. Results: The segmentation of blood vessels is implemented using the U-Net model, the sensitivity and the specificity are measured as 97% and 88% respectively. The accuracy and the F1 score have been measured as 96% and 98% respectively. Hence, the proposed deep learning architecture with hybrid PolyNet models is implemented for OD and OC separately to produce an appropriate result. By increasing the number of layers, more features are identified. The current work deals with the retinal fundus images and achieves the highest accuracy of 96.21% in the ACRIMA dataset with sensitivity and specificity of about 97.32% and 94.26% respectively. Conclusion: Blood vessel segmentation is achieved with U-Net and the segmentation of the optic disc and the optic curve have been achieved and the severity level of glaucoma has been identified using hybrid PolyNet with high accuracy.