Domain Adaptation Using DCGAN For Glaucoma Diagnosis

The second most prevalent eye condition is glaucoma to cause neurodegenerative sickness among eye conditions. Improper intraocular pressure in the human eye is the main reported reason for this disorder. The beginning stages of glaucoma do not exhibit any symptoms. Glaucoma patients show no symptoms and no signs of illness early in the disease, but vision loss occurs as the disease advances. Techniques for glaucoma auto-detection can be advanced using computational methods from machine learning in conjunction with digital image processing. Domain shifts are very common in fundus images as they are captured on different devices with different rules. The domain shift is a very pressing issue in glaucoma diagnosis. In this paper, we employ a Deep Convolutional Generative Adversarial Network (DCGAN) to generate source-like target images which are tested on a classification model, ResNet-50. The results show that we can successfully achieve good accuracy using domain adaptation.