Glaucoma Detection Using Multiple Feature Set With Recurrent Neural Network

Abstract One of the most widespread illnesses of blindness is glaucoma. Optic nerve are essentialfor clear vision, but glaucoma effects the optic nerves and results blurred vision. This condition is often exacerbated by abnormally high intra-ocular pressure. Accurate early identification and continuous screening can help to minimize loss of vision. A non-invasive computer-aided diagnosis treatment uses optical fundus images to detect glaucoma in its early stages. This work includes image preprocessing, optic disk (OD) segmentation, feature extraction from the OD and recurrent neural network classification to identify glaucoma. The performance of the proposed system is tested using fundus image datasets such as DRISHTI-GS and Large-Scale Attention-Based Glaucoma (LAG). By this method, glaucoma detection accuracyof 96.1% is obtained for DRISHTI-GS and 92.73% for LAG dataset, which is higher thanthe existing state of arts. Proposed procedure can help ophthalmologists diagnose glaucomawith good performance.