This dataset contains 1,000 retinal fundus photographs captured using a CrystalVue fundus camera and an accompanying CSV file with binary labels for a wide range of ophthalmic conditions. Each column represents a specific disease (e.g., Microaneurysms, Hard Exudate, Glaucoma) and a diabetic retinopathy grade (dr_level) following the ICDR scale. The data is intended for researchers, data scientists, and clinicians developing or evaluating computer-vision models for multi-label eye disease detection. Typical use cases include training deep-learning classifiers, benchmarking algorithms, and teaching medical imaging techniques.
The dataset enables training of deep-learning models that automatically identify diabetic retinopathy and other retinal pathologies from fundus photographs. Researchers can benchmark multi-label classification algorithms, develop screening applications for low-resource eye care, and create educational resources for teaching medical image analysis. Public health analysts may use the data to explore disease prevalence patterns, while software engineers can prototype end-to-end pipelines for automated eye-exam interpretation.
Classes such as rare retinal disorders are sparsely represented, leading to potential class-imbalance issues during model training. The sample size of 1,000 images may limit generalization to broader populations. The dataset lacks longitudinal follow-up and contains only static annotations, so progression analysis is not possible.
© 2026 - Copyright AIKosh. All rights reserved.