Cataract, a common age-related eye condition, can lead to blindness if left untreated. This project focuses on enhancing the accuracy and efficiency of cataract detection. A MEDNet-based model is trained on imbalanced eye image data, leveraging latent vectors and sampling techniques for data balance. Cross-validation reveals superior accuracy and speed compared to existing methods. This work advances early and accurate cataract detection, potentially serving as an automated tool to alleviate the burden on ophthalmologists and improve eye care quality.