A CNN-based hybrid model to detect glaucoma disease

Glaucoma is an eye disease caused by damage to the optic nerves and is a common cause of incurable blindness worldwide. If glaucoma is diagnosed early, vision loss can be prevented with regular exams and treatment. If diagnosed too late, the disease can cause severe damage to the optic nerve that cannot be reversed, leading to loss of central vision and blindness. Therefore, early diagnosis of the disease is critical. Most of the studies conducted in recent years have presented Deep Learning (DL) based architectures that use an automatic computerized system based on segmentation of hand-made features in fundus images. In this study, we seek to help experts detect glaucoma using a model that combines Deep Learning and Machine Learning using raw fundus images. Deep features are extracted using a new Convolutional Neural Networks (CNN) model. Deep features have been used in popular traditional Machine Learning methods (ML) for classification such as Adaboost, k Nearest Neighbor (kNN), Random Forest (RF), Multilayer Perceptron (MLP), Support Vector Machines (SVM), and Naive Bayes (NB). The performances of the hybrid models were evaluated using the ACRIMA dataset of 705 images. The dataset is reserved for 80% training and 20% testing data. Experimental results show that the hybrid model of CNN and Adaboost has the highest success rate with 92.96% accuracy, 93.75% F1 score and an AUC value of 0.928.