Performance Evaluation of Hybrid and CNN Models for Glaucoma In vestigation on the ACRIMA, REFUGE, and G1020 Dataset

Glaucoma, a leading cause of permanent blindness worldwide, is caused by damage to the optic nerve caused by high intraocular pressure. Preventing vision loss requires early identification and surveillance. Recent studies have used algorithms to analyze segmentation and deep learning of retinal images. Deep learning, more specifically Convolutional Neural Networks (CNNs), revolutionize this discipline and automate the diagnosis of glaucoma. CNNs are quite good at extracting features; they can identify minute structural alterations surrounding the optic nerve head. Differences in the symptoms between the patient groups, however, continue to pose difficult problems for the generalization of deep learning models. It aims to analyze the effectiveness of a hybrid deep learning strategy in comparison with conventional methods for glaucoma screening. CNN models ResNet, DenseNet, Inception, and hybrids are adopted for training three glaucoma datasets, namely Acrima, GS1020, and Refuge.