Detection Glaucoma using Deep Learning Algorithms

A common eye disorder called glaucoma may result in irreparable vision loss if it is not identified early on. Automated methods that use fundus images have become a potential tool to help physicians diagnose patients more accurately and quickly. This research acquired a wide dataset that included both glaucomatous and non-glaucomatous patients. A range of deep learning and machine learning models were developed and refined for the purpose of detecting and classifying glaucoma. Sensitivity, specificity, accuracy, and AUC-ROC were among the performance assessment criteria utilized to evaluate the models. According to findings, convolutional neural networks (CNNs), a kind of deep learning model, perform better than other deep learning algorithms in terms of accuracy and clinical integration potential. The results of this study advance the area of automated glaucoma evaluation and provide relevant information to academics and clinicians who are trying to improve the diagnosis and treatment of glaucoma. The results of our study demonstrate that the Bilateral U-Net model achieves superior performance across all tested datasets, with the highest accuracy of 92.40% on DRISHTI-GS, 91.60% on ORIGA, and 90.70% on ACRIMA, significantly outperforming other state-of-the-art models such as DE-NET, ResNet-50, and Inception V3. Additionally, the DRISHTI-GS dataset is identified as the most prominent in terms of performance, achieving the highest accuracy, precision, sensitivity, and specificity across all evaluation metrics Furthermore, the potential for the integration of multiple models to enhance the accuracy and reliability of glaucoma assessment is emphasized, thereby facilitating future developments in this field.