ABSTRACT
The glaucoma is a retinal disease that causes permanent blindness if not detected earlier. The asymptotic nature of glaucoma is a big hurdle in the early-stage diagnosis. The increased intraocular pressure (IOP) normally greater than 22 mmHg affects the retinal ganglion cells, and defects of glaucomatous visual field and large diameter of optic cup (OC) in retinal optic disk (OD) are the main reasons for the glaucoma occurrence. The glaucoma can be diagnosed by calculating optical cup-to-disk ratio, IOP measurement, optic nerve head (ONH) evaluation and retinal nerve fiber layer (RNFL) thickness. This chapter focuses on the machine learning and the deep learning approaches of the recent years for the glaucoma diagnosis and detection at an earlier stage. Different methodologies of published studies in machine learning of the last 10 years and in deep learning of recent 7 years, datasets used in these studies and their performance have been evaluated. The publications with high impact factor including both the OC and disk segmentation and the classification techniques have been focused. The machine learning approaches require the fabricated features for the classification of glaucoma. The best feature vector diagnosis the glaucoma with greater accuracy. The fundus images are globally used by the ophthalmologists and clinicians for glaucoma diagnosis, which clearly shows the morphological features of the OD leading to glaucoma. The optical coherence tomography (OCT) images are also being used by the researchers for the glaucoma diagnosis in earlier stages. Recently deep learning has gained a significance over machine learning due to its automatic learning capability for developing intelligent systems for early diagnosis of glaucoma. But the high computation resources and large datasets for training are required by the deep learning models. It can be accomplished by the data augmentation and pretrained networks. Based on all the studies, it is needed to design such automatic systems based on deep learning models that can diagnosis glaucoma in early stages with greater accuracy due to its severity and high occurrence and can be improved over time by using novel methods for screening, diagnosis and detection of glaucoma. This will minimize the need of skilled and trained clinicians required for the glaucoma diagnosis. More the large datasets of retinal fundus and the OCT images are also required in the deep learning to build efficient and reliable glaucoma diagnosis systems. The glaucoma diagnosis in early stage through intelligent systems will minimize the risk of suffering in permanent blindness.
