Retinal Fundus Image Classification Using Generative Adversarial Networks

This study presents an innovative method for classifying retinal fundus images using a novel combination of Generative Adversarial Networks (GANs) and Inception V3. This unique method has demonstrated impressive results, achieving an 87.44% of test accuracy. The proposed model also excels in precision (87.25%), recall (87.0%), and F1-score (86.75%) and has an ROC value of 0.9775. To address the overfitting and underfitting issues, this work uses K-fold cross-validation, ensuring the robustness and generalizability of the model. This work contributes to the field of eye health, mainly in image generation and classification of retinal images, and combines the GANs model with the transfer learning model Inception V3 for medical image analysis.