Glaucoma is a disease that damages the optic nerve, which is responsible for sending visual images from eye to brain. Glaucoma occurs when the fluid in the eye does not drain as in the normal conditions. In such a situation, the fluid builds up and increases pressure in the eye gradually, which will damage the optic nerve. World Health Organization (WHO) has estimated that over 64 million people living with glaucoma in the world in 2019, where more than 60% of glaucoma patients (51 million) live in Asia countries. Also, it is estimated that by 2030 over 95 million people will have glaucoma]. Glaucoma is identified as the second leading cause for blindness worldwide. Glaucoma detection is required for preventing from blindness and reducing cost for surgeries such as trabeculectomy, tube shunt implantation for socially as well as economically. Blindness from glaucoma can often be prevented with early treatment. Treating glaucoma will not bring back any vision lost, but it can help save the current sight. Thus, it is important to address the problem of glaucoma diagnosis.
Addressing Computer-Aided Diagnosis (CAD) of ocular diseases is an active research area due to the benefits such as generating accurate image classification results and providing tool support for the glaucoma diagnosis process. Most of the existing studies have used machine learning techniques focusing on only one dataset and obtained reasonable classification accuracy. In this study, we consider different datasets to achieve better classification accuracy compared to the existing studies. The proposed research applies different deep learning and computer vision-based techniques to detect various ocular diseases like glaucoma for different types of retinal fundus images. Moreover, we apply parameter tuning and different optimizers to increase the classification accuracy of the learning models. Therefore, improvement of the classification accuracy and avoiding overfitting of machine learning techniques can be considered as the novelty of this research.
Further, we apply explainable AI techniques to provide visual explanations for the predictions based on the underline deep learning model. This framework with segmentation, classification and explainability is deployed as a web application. Such computational decision support will be beneficial for the professionals in ophthalmology such as optometrists and ophthalmologists.