JMIR Publications

ABSTRACT

Background:

Glaucoma means irreversible blindness. Globally, it is the second retinal disease leading to blindness, just preceded by the cataract. Therefore, there is a great need to avoid the silent growth of such disease using the recently developed Generative Adversarial Networks(GANs).

Objective:

This paper aims to introduce GAN technology for the diagnosis of eye disorders, particularly glaucoma. This paper illustrates deep adversarial learning as a potential diagnostic tool and the challenges involved in its implementation. This study describes and analyzes many of the pitfalls and problems that researchers will need to overcome in order to implement this kind of technology.

Methods:

To organize this review comprehensively, we used the keywords: ("Glaucoma", "optic disc", "blood vessels") and ("receptive field", "loss function", "GAN", "Generative Adversarial Network", "Deep learning", "CNN", "convolutional neural network" OR encoder), in different variations to gather all the relevant articles from five highly reputed databases: IEEE Xplore, Web of Science, Scopus, Science Direct, and Pubmed. These libraries broadly cover technical and medical literature. For the latest five years of publications, we only included those within that period. Researchers who used OCT or visual fields in their work were excluded. However, papers that used 2D images were included. A large-scale systematic analysis was performed, then a summary was generated. The study was conducted between March 2020 and November 2020.

Results:

We found 59 articles after a comprehensive survey of the literature. Among 59 articles, 29 present actual attempts to synthesize images and provide accurate segmentation/classification using single/multiple landmarks or share certain experiences. Twenty-nine journal articles discuss recent advances in generative adversarial networks, practical experiments, and analytical studies of retinal disease.

Conclusions:

Recent deep learning technique, namely generative adversarial network, has shown encouraging retinal disease detection performance. Although this methodology involves an extensive computing budget and optimization process, it saturates the greedy nature of deep learning techniques by synthesizing images and solves major medical issues. There is no existing systematic review paper on retinal disease utilizing generative adversarial networks to the extent of our knowledge. Two paper sets were reported; the first involves surveys on the recent development of GANs or overviews of papers reported in the literature applying machine learning techniques on retinal diseases. While in the second group, researchers have sought to establish and enhance the detection process through generating as real as possible synthetic images with the assistance of GANs. This paper contributes to this research field by offering a thorough analysis of existing works, highlighting current limitations, and suggesting alternatives to support other researchers and participants to improve further and strengthen future work. Finally, the new directions of this research have been identified.