Detection of Glaucoma Using A CNN-Based Hybrid Model with Color Fundus Images

Glaucoma is a disease of the eye that damages the optic nerve, often as a result of fluid buildup in the front part of the eye. Elevated intraocular pressure due to this accretion may cause damage to the optic nerve and result in loss of vision. It may start peripherally and then progress to the point of interfering with central vision. Patients with glaucoma typically do not have noticeable symptoms until many months or years after the onset of optic nerve damage. This article will present a new trend in researching DL architectures in automated systems to segment handcrafted features in fundus images. In this novel approach, the model integrates DL and ML techniques, applied to raw fundus images, to help experts detect glaucoma. Outputs of the CNN were taken as deep features and forwarded through different machine learning algorithms such as XGBoost, logistic regression, linear regression, Adaboost, stacking, and linear discriminant analysis for the classification of images. The performances of these hybrid models were tested with the ACRIMA dataset, which consists of 705 images divided by 70% for training purposes and 30% for testing purposes. Experimental results show that the highest degree of success is achieved with the hybrid model combining CNN with XGBoost, which achieves 96% accuracy.