A Deep Learning-based Automatic Method for Early Detection of the Glaucoma using Fundus Images
Ayesha Shoukat, Shahzad Akbar, Khadija Safdar
Abstract
Glaucoma is a retinal abnormality that affects the ONH (optic nerve head) and results in loss of vision if it is diagnosed at an advanced stage. It does not show the symptoms at the initial stage due to its asymptotic nature. Timely detection and preventive treatment can save the everlasting blindness caused by glaucoma. In this paper, a method using the deep learning model, the convolutional neural network architecture is proposed to develop the glaucoma diagnosis system at the initial stage. Two datasets of retinal fundus images are used named DRISHTI-GS and G1020 for the classification of the glaucomatous and the healthy images. Multiple filters are applied to the fundus images to provide quality images for the classification. The EfficientNet architecture is used in the proposed model for the classification to detect glaucoma at the early stage. The proposed method has obtained the state of the art results with an accuracy of 98%, sensitivity of 95.19% and specificity of 94% on the DRISHTI-GS dataset. The presented model will help the clinical diagnostic system to make reliable decisions about the diagnosis of early-stage glaucoma.