Abstract:
Glaucoma is an eye problem that affects the retina and weakens the nerve cells that assist in visual recognition. This disease can be the cause of blindness and visual impairments if not treated in time. Increase in Intraocular pressure (IOP) is the main cause for glaucoma presence and succession. The diameter of the optic cup within optic disc region is increased due to the excessive retinal pressure. Cup to disc ratio (CDR) is the mainly used clinical feature to diagnose glaucoma. The increment in CDR is the main factor to classify glaucoma patients. This causes the retina to be affected and also decrease the retinal thickness. For glaucoma diagnosis, CDR is calculated by two commonly used imaging modalities i.e., OCT (Optical Coherence Tomography) and Fundus Imaging. These two imaging technologies based tools are of great use to identify/monitor retinal tissue structures to treat earlier with ocular diseases. Previously different researches have computed CDR in fundus retinal images by segmenting optic cup and disc regions and then their diameters calculation. This research focuses on reliable extraction of CDR using optical coherence tomography, and then classifies the subject as glaucomatous or normal. A new method for extraction of inner limiting membrane (ILM) and retinal pigment epithelium (RPE) layers from OCT scans. The proposed method then used both ILM and RPE for calculation of cup and disc which eventually help in CDR calculation. Based on calculated CDRs, the system classifies the patient as normal or glaucomatous. The evaluation is done using a local dataset of 50 optic nerve head centered OCT scans and results show the validity of proposed system. The dataset is obtained from Armed Forces Institute of Ophthalmology (AFIO), Pakistan. For clinical correlation, computed CDRs on whole dataset are compared with annotated CDRs by four ophthalmologists. CDR values greater or equal than 0.5 are considered glaucomatous, and less than 0.5 as normal. The proposed system provides a mean error on computed CDR when compared with all annotated CDR on whole 50 dataset to be 0.138 ± 0.026. The proposed system has shown excellent results with 87% sensitivity, 79% accuracy and 72% specificity on AFIO dataset when correlated annotated and CCGV values.
