Enhancing Glaucoma Diagnosis with a Novel Optic Disk Localization and Classification Framework Based on U-Net and ResNet-50

Glaucoma is one of the chronic eye ailments that result in irreversible visual deterioration and is widely recognized as a significant contributor to the phenomenon of blindness on global terms. Early detection will help in drastically reducing the course of the disease. The paper proposes a three-phase novel framework for the localization of optic discs and glaucoma detection. It concentrated on early-stage identification to improve treatment outcomes. First, the median filtering and histogram equalization constitute the techniques for that process. With these techniques, different kinds of noise, such as Rayleigh, salt and pepper, Gaussian, and gamma noise, can efficiently be reduced in fundus images. At the result of this technique, the clarity of the images gets improved to a large extent and also makes the further analysis more accurate. In the next stage, location and segmentation of the optic cup and optic disc will be done accurately to obtain the desired output. To get the desired outcome, the optic disc and cup will be precisely located and segmented in the stage that comes next. For this purpose, use is made of the U-Net architecture in delineating these critical structures that lie within the eye. In the last stage, fundus images are classified to be either glaucomatous or non-glaucomatous using transfer learning-based models, prominently ResNet-50. With ResNet-50, the overall capability of the model has been highly enhanced to successfully analyze and identify complex patterns inside retinal images. Experimental results of this study, conducted using the ORIGA dataset, further testify to the effectiveness of the suggested framework. An accuracy rate of 96.78% is attained by the proposed framework, which turned out superior to the state-of-the-art approaches.