الفهرس | Only 14 pages are availabe for public view |
Abstract Glaucoma disease (GD) is a rapidly growing consequence of Glaucoma estimates globally. The importance of accurate GD diagnosis in improving patient care and treatment outcomes has shown a significant increase in research interest in recent years. The advancement in Deep Learning (DL) approaches has proven to be superior to traditional detection methods. This thesis proposes a deep learning construction using a convolutional neural network (CNN) for the automated detection of glaucoma from fundus images to distinguish between Glaucoma-affected and healthy images. The important features from the input data are extracted using CNN that has been built and trained. On the ACRIMA dataset, which contains a total of 705 images, these frameworks have been tested and trained. The proposed CNN-based system was compared with other pre-trained models (EfficientNetB0, ResNet101, Resnet50, VGG16, and InceptionV3). The aim of this thesis is to compare the performance obtained from different configurations with CNN architectures and hyper-parameter tuning. Among the considered deep learning models, the EfficientNetB0 model showed the highest accuracy of 98% for the ACRIMA fundus image dataset. The best performance was obtained using the proposed CNN network, achieving 98.1% accuracy, 98.31% sensitivity, and 97.85% specificity. Additionally, this study presents a comparative analysis of how changes in the hyper-parameter of the model can affect classification performances. As a result, the proposed method can be used as an effective diagnostic tool for Glaucoma detection. |