Enhanced Effective Baseline Model for Glaucoma Prediction Using U-Net and Hyperparameter-Tuned CNN with CLAHE

Abstract

Glaucoma is a progressive eye disease that can lead to permanent vision loss when diagnosis is delayed. Our study proposes an enhanced effective baseline framework for automated glaucoma prediction from retinal fundus images by integrating image enhancement, segmentation, deep feature extraction, and machine learning classification respectively. The methods Contrast limited adaptive histogram equalization is applied to improve image contrast and reveal important retinal structures. The U-Net architecture is then employed to segment the optic disc region and reduce the influence of irrelevant retinal areas respectively. The segmented regions are subsequently processed using a hyper parameter-tuned Convolutional Neural Network to learn discriminative deep features efficiently. The optimized features are provided to an enhanced support vector machine classifier for glaucoma prediction. The hyper parameters of the CNN are systematically optimized to obtain more representative features and improve classification performance effectively. The incorporation of contrast enhancement, U-Net segmentation, optimized deep feature learning, and Enhanced Support Vector Machine provides an effective framework for automated glaucoma screening and early identification of glaucomatous changes respectively.