The Detection of Glaucoma in Fundus Images Based on Convolutional Neural Network

Glaucoma is a common disease affecting the human retina, primarily caused by elevated intraocular pressure. Early intervention is crucial to prevent damage to the affected organs, which could lead to their dysfunction. This paper focuses on enhance diagnosis accuracy of the system to determine if a patient is at risk of developing glaucoma. In this paper a novel convolutional neural network (CNN) designed, specifically for the detection of glaucoma in fundus images. This architecture optimizes for the unique characteristics of fundus imagery, enhancing detection accuracy, and also compiled a large and diverse dataset of fundus images, crucial for training and validating our CNN model. The dataset includes a significant number of images with detailed annotations, ensuring robust model training. In addition, implemented sophisticated image preprocessing methods to enhance the quality of the fundus images. These techniques, including noise reduction and contrast enhancement, significantly improve the input data quality for the CNN. The system operates in three stages. First, it preprocesses the image by cropping, enhancing, and resizing it to a consistent 256×256 pixels. Next, it employs an advanced feature extraction to analyses key features of the optic disc and optic cup in retinal images. Finally, the Soft-Max function classifies the images, identifying those with glaucoma and distinguishing them from normal eye samples. The model's performance was thoroughly evaluated using various metrics like accuracy, Sensitivity, specificity, and the area under the curve are metrics used to evaluate the performance of a diagnostic test. Sensitivity measures the test's ability to correctly identify positive cases, specificity assesses its accuracy in identifying negative cases, and the area under the curve indicates the overall effectiveness of the test across different thresholds. The results achieved by the proposed system were thoroughly analyzed, revealing a high accuracy rate in glaucoma classification, reaching 99%.