Automated Glaucoma Detection and Assessment using CNN from Fundus Images

Glaucoma is a widespread eye illness that is the second most frequent cause of vision impairment worldwide. Glaucoma causes slow and irreversible eyesight loss by damaging the optic nerve, making it a complex ailment. The illness is often associated with a buildup of pressure inside the eye. Texture features derived from fundus images are used to classify and automate the identification of glaucoma. Detecting and extracting specific textural features involves using the energy distribution in wavelet subbands. The study uses the Daubechies wavelet filter to capture energy features from the image using the 2-D discrete wavelet transform. The sequential forward feature selection strategy determines the most essential features. The methodology utilises a dataset of 5504 fundus images, evenly split between glaucomatous and normal instances, with 2752 images in each group. The classifiers, including the Support Vector Machine, Artificial Neural Network, and Convolutional Neural Network, utilise the best attributes to categorise the images as normal or abnormal. Analysing experimental results assesses the computational accuracy of the classifier.