Advanced Techniques in Automated Glaucoma Diagnosis: A Comprehensive Survey of Deep Learning and Image Processing Approaches

Abstract

Optic neuropathy, like glaucoma, is progressive, but it is still the second leading cause of blindness in the world and this makes early, precise and automated diagnosis is very much essential. Recently, there has been tremendous advancements in deep learning & image processing technique and making it possible to detect glaucoma by retinal fundus image. In this review we highlight recently introduced state-of-the-art methods for automated glaucoma diagnosis focusing on the use of convolutional neural network (CNN), transfer learning as well as conventional image based feature extraction techniques. Wavelet, Gabor and curvelet based methodology, Retinex based enhancement, segmentation based analysis and hybrid machine learning models have been applied to determine the structural and textural changes in optic nerve head. The review identifies also emerging challenges, such as the collection of large scale diverse data for better generalization, and more powerful statistical evaluation approaches Expected challenges are discussed these include the acquisition of large and diverse data sets to enhance generalization across different imaging conditions and powerful statistical evaluation methods. It also debates whether detection in the early stages of glaucoma can be achieved and discuss what these models are explainable to look for in a diagnosis. To conclude, this review summarizes current techniques for automated glaucoma detection and highlights potential gaps in for future research.