Empirical Evaluation using Generative AI models for the Multiclass Classification of Retinal Diseases Images

In today’s era the robust and reliable models of Generative AI (Gen AI) are used in different research domains. In this research we have used Super Resolution Generative Adversarial Networks (SRGAN) model in healthcare domain specifically retinal images of eyes. The model performs multiclass classification on the four different diseases of eye - Choroidal Neovascularization (CNV), DRUSEN, Diabetic Macular Edema (DME), and NORMAL form and predicts their occurrence. The standard dataset from Mendley is used for research. The training images in each class are augmented by using synthetic images generated from SRGAN. The generator of SRGAN has generated 2000 synthetic images of each class and added them to the training dataset. Thereafter, a custom CNN model is developed for multiclass classification. The model has outperformed the existing approaches and resulted in macro average and weighted average of 99% each respectively.