GAN and ResNet Fusion A Novel Approach to Ophthalmic Image Analysis for Glaucoma

Abstract

Glaucoma is a major cause of blindness, often undetected in early stages due to lack of symptoms. Addressing this, research study developed a deep learning framework integrating Generative Adversarial Networks (GANs) with Residual Neural Networks (ResNet) to enhance glaucoma detection from fundus images. Utilizing GANs for data augmentation, we enriched the training set with synthetic images that improve feature recognition, while ResNet, fine-tuned on this data, performed high-precision classification. The GAN's discriminator, trained using binary cross-entropy loss, concentrating to extract key indicators of glaucoma from these fundus images, with its performance assessed by its accuracy in distinguishing real from synthetic images. The GAN-ResNet channel exploited the discriminator's feature extraction coupled with ResNet's deep learning capabilities to classify the fundus images with refined accuracy. The proposed model final layer is fine-tuned for binary classification between glaucomatous and healthy images, with the loss function modified for medical dataset imbalances. Through wide testing, the GAN-ResNet model proven a remarkable 98% accuracy in analysing glaucoma, showing high predictive results. This validates that the proposed model is helpful in detecting glaucoma early. It highlights how well-advanced neural networks work for analysing medical images.