Early Glaucoma Detection using LSTM-CNN integrated with Multi Class SVM

Abstract

Glaucoma, a progressive eye disease, is a major public concern on health due to its gradual onset and the possibility of irreversible vision loss. Early glaucoma detection is critical because it allows for timely intervention and management, lowering the risk of severe visual impairment. To address this pressing need, we present a comprehensive glaucoma detection methodology that focuses on image processing techniques and machine learning models. The initialization and preprocessing of retinal fundus images obtained from the DRIVE database is the first step in our approach. These images are resized to a standard size, grayscaled, and blurred with Gaussian blur to ensure consistency and noise reduction. Our methodology is built around feature extraction and modeling. We harness the power of deep learning, specifically Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNNs), which we integrate seamlessly with multi-class Support Vector Machines (SVMs). This synergy enables our Deep Flex SVM-MC model to capture intricate data patterns during training while also demonstrating exceptional adaptability in multi-class classification tasks. The proposed model has a glaucoma detection accuracy of 97.2%, an exceptional sensitivity of 97.53%, indicating its proficiency in correctly identifying glaucoma cases, and a specificity of 96.4%.