Hybrid features and optimization-driven recurrent neural network for glaucoma detection

Glaucoma is considered as the main source of irrevocable loss of vision. The earlier diagnosis of glaucoma is essential to provide earlier treatment and to reduce vision loss. The fundus images are transfigured in the ophthalmology and are used to visualize the structures of the optic disc. However, accuracy is considered as a major constraint. To increase accuracy, an effective optimization-driven classifier is developed for glaucoma detection. The proposed Jaya-chicken swarm optimization (Jaya-CSO) is employed for training the recurrent neural network (RNN) for glaucoma detection. The proposed Jaya-CSO is designed by integrating the Jaya algorithm with the chicken swarm optimization (CSO) technique for tuning the weights of the RNN classifier. The method utilized optic disc features, statistical features, and blood vessel features for the determination of the glaucomatous region. The features obtained from the optic disc, blood vessels, and the fundus image is formulated as a feature vector. Finally, the glaucoma classification is done using RNN using the feature vector such that the RNN is trained using the proposed Jaya-CSO. The proposed Jaya-CSO outperformed other existing models with maximal accuracy of 0.97, the specificity of 0.97, and sensitivity of 0.97, respectively.