Experimental Approach to Identify the Optimal Deep CNN Models to Early Detection of Glaucoma from Fundus CT-Scan Images

Glaucoma is an eye disorder that causes the optic nerve to be affected in glaucoma cause of improved Intraocular fluid Pressure (IOP) in the eye. The only way to prevent blindness is to detect glaucoma early, which can only be done through regular eye examinations. Additionally, earlier diagnosis of glaucoma is challenging to determine and requires several costly vision examination protocols and skillful ophthalmologists. The proposed work’s goal is to detect Glaucoma using Convolution Neural Networks (CNNs) to speed up the diagnosis process for timely prevention. CNNs have proven effective in discovering even subtle changes produced by glaucoma. The experimental study using three different Deep Learning (DL) algorithms CNN, MobileNet, and ResNet are carried out to analyses the best model. All these models are pretrained on the image data generator for detecting and classifying glaucoma from the fundus photographs. A Gaussian filter is applied in order to remove noise from the images. Three different datasets ORIGA, SCES, and G1020, which contains images of both healthy and advanced glaucoma, are used to train the model. Also, these databases contain colored images of retina and retinal fundus images that are used for testing, and screening the glaucoma. The proposed method attained the best accuracy of 95% from the ResNet-50 algorithm. Also, each suggested model’s performance has been evaluated with the results of each CNN model. Additionally, a custom web application has been developed with CNN-based models to classify Glaucoma. The outcomes show that the CNN algorithms have the ability to develop into a time-saving and cost-efficient tool for analyzing populations intended to prevent irreparable blindness caused by glaucoma.