ABSTRACT
Age-related macular degeneration is the leading threat to older adults in terms of their visual health, and its effective management is tied to early detection. Traditional diagnostic methods do not work well when it comes to the accurate classification of AMD stages. The deep learning algorithms, though promising, face problems of an imbalanced dataset and limited advanced AMD cases. We present a hybrid deep learning model herein that combines Vision Transformers' global contextual extraction, EfficientNet's local feature capture capability, and the synthetic data augmentation capability of GANs. Integrating these approaches improves accuracy and generalization by overcoming class imbalance and limited data, offering more reliable and robust AMD classification that outperforms conventional CNN-based techniques for clinical diagnosis.
