Leveraging CNN and Fundus Imaging for Enhanced Glaucoma Detection

Glaucoma, the foremost contributor to permanent vision impairment worldwide, necessitates early and accurate detection methods to prevent vision loss. Traditional diagnostic approaches often rely on subjective analysis by specialists, which can be time-consuming and prone to variability. This is prompting researchers to sort out the issue using some automated detection methods. This work proposes a 9-layer CNN for automatic glaucoma classification. The diagnosis of glaucoma is made using 1692 fundus images from four labeled databases, which included 953 normal and 739 glaucomatous images. A 9-layer CNN is trained on this dataset to extract essential characteristics, that are identified as normal or glaucoma during testing. The suggested technique is trained and evaluated with TensorFlow and Python Jupyter Notebook, and the system is implemented using an NVIDIA Quadro P2000 GPU. To assess the model’s efficacy, 80% of the dataset undergo training and 20% testing. The suggested CNN yielded 99% training accuracy and 98.6% testing accuracy with ACRIMA dataset. These findings show that this deep learning model exhibits excellent performance in detecting glaucoma from fundus images, and they imply that the proposed approach can assist ophthalmologists in accurately diagnosing glaucoma.