Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{189910,
        author = {Nandini Rathod and Shital Zalke and Vanshika Bahadure and Sahil Khobragrade and Yashkumar Baghele},
        title = {AI-Powered Retinal Disease Classifier Using Fundus Images},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {8},
        pages = {2606-2611},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=189910},
        abstract = {Retinal pathologies like diabetic retinopathy, cataract, and glaucoma constitute the most prevalent causes of avoidable blindness globally. Specialist ophthalmologist access continues to be a rarity, especially in areas where resources are scarce, even though the detection needs to be done early. To automatically identify and classify the four classes—Cataract, Diabetic Retinopathy, Glaucoma, and Normal—a fundus image-based AI-powered retinal disease classifier is proposed in this study. With an optimized pre-trained ResNet-18 model for multi-class classification, it utilizes transfer learning. To improve generality, the hand-collected dataset was preprocessed with augmentation, normalization, and resizing methods. Stratified splitting was used for balanced training (70%), validation (15%), and test (15%) sets. With overall accuracy of 90%, the model was optimized across three epochs using Adam optimizer and cross-entropy loss. Performance testing under diabetic retinopathy (0.97) and cataracts (F1-score: 0.91) included exceptionally high F1-score and recall; detection of glaucoma, however, was tough (F1-score: 0.84). From confusion matrix analysis, it was revealed that the majority of the misclassifications between glaucoma and normal ones reflected higher diversity in the dataset along with more sophisticated feature extraction. The model was implemented as a web application based on Flask for improving clinical utility with real-time prediction in a user-friendly interface even accessible to non-professionals. The approach has widespread scope for mass retinal screening, especially in telemedicine and rural medicine. Though the initial results are encouraging, glaucoma detection needs to be improved, diversity of datasets should be increased, and explain ability should be ensured. To alleviate the worldwide burden of preventable blindness, this study helps in the development of affordable AI-powered ophthalmology solutions.},
        keywords = {(CNN), Fundus Imaging, Retinal Disease Classification, Diabetic Retinopathy, Cataract, Glaucoma, ResNet-18, Transfer Learning, Medical Image Analysis, Telemedicine, Flask Web Application, Automated Screening, Preventable Blindness.},
        month = {January},
        }