Glaucoma Detection Approach based on Colour Fundus Pictures

Glaucoma is a common retinal disorder that has an impact on the optic nerve, resulting in irreversible sight loss if left untreated. Although early detection is crucial for optimal management, manual detection is difficult and needs highly competent ophthalmologists. To overcome this issue, this study Employ Convolutional Neural Networks (CNNs) and further deep learning techniques, to detect glaucoma automatically. The study employs a standardization technique on a dataset of 5000 retinal pictures to ensure that all images are uniform at $256\times 256$ pixels. Data augmentation techniques such as rescaling, rotation, vertical and horizontal shifts are used consistently to improve dataset variety and model resilience. The MobileNetV3 architecture is used for glaucoma detection, and two optimizers, Stochastic Gradient Descent (SGD) and Adam are used during model training. Pre-processing refines incoming pictures further by turning them to greyscale, adding Gaussian noise, and ensuring pixel values are inside the [0, 255] range. Accuracy, precision, recall, specificity, and F-measure are all part of performance evaluation. Comparative analyses evaluate the suggested model’s efficacy, assuring its dependability and quality. The results show that the model is proficient in detecting glaucoma due to its excellent classification accuracy. This work improves automated glaucoma detection, perhaps assisting in early diagnosis and treatment, so protecting patients from irreparable vision loss caused by this “silent thief of vision.”