Authors: Tukaram C. Bhoye, S. V. Gumaste
Abstract: Eye diseases such as diabetic retinopathy, glaucoma, and cataract are among the leading causes of visual impairment worldwide. Early diagnosis is crucial to prevent irreversible vision loss, particularly in areas with limited access to specialized ophthalmic care. Recent progress in deep learning (DL) has enabled automated analysis of retinal fundus images, facilitating faster and more accurate detection of eye-related disorders. This paper presents a comprehensive review of deep learning-based approaches for multi-disease eye detection and classification using fundus images. Various techniques, including convolutional neural networks (CNNs), transfer learning methods, transformer-based architectures, and hybrid models, are examined. The study also reviews commonly used retinal datasets, evaluation metrics, and comparative performance of different approaches. In addition, key challenges such as dataset imbalance, limited generalization, and lack of model interpretability are discussed. Future research directions, including the use of explainable AI, lightweight mod-els for edge deployment, and privacy-preserving techniques, are highlighted. The findings of this review provide valuable insights into current developments and support the ad-vancement of reliable and clinically applicable AI-based ophthalmic diagnostic systems.
DOI: http://doi.org/10.5281/zenodo.20848646
