Abstract:
Glaucoma is a neurodegenerative eye disease which occurs due to the elevated IOP
(Interocular Pressure) that builds in the eye. The aqueous humour fluid in the cham ber of eye presses the visual nerve due to which the condition of optic nerve deterio rates causing permanent blindness. Traditional methods such as gonioscopy, tonom etry, dilated eye exam for the diagnosis is time consuming, expensive and require
multiple visits to the hospital. Due to this, detection at early stage is compromised.
With the advent in deep learning due to its pattern recognition capabilities, it is
quite commonly used in early prediction of ocular diseases, glaucoma being one of
them. This research focuses on developing a new methodology through which glau coma can be detected using coloured fundus retinal images. The system consists of
two basic modules: first segmenting the optic disc using a segmentation model i.e.
customised Segnet being followed by a classification architecture that consists of an
ensemble of VGG-16 and a customized CNN. The results have shown that the pro posed model provides improved accuracy along with better sensitivity, specificity as
compared to other existing architectures giving better generalization results in less
time using less memory and computationally intensive system aids in early detection.
