Ophthalmology Datasets — Fundus & OCT Retinal Data

Ophthalmology imaging datasets capture the retina, optic nerve, and macula to power clinical AI for screening, diagnosis, and disease monitoring, and they are central to scalable detection of the leading causes of preventable blindness. The two dominant modalities are fundus photography and optical coherence tomography. Color fundus photographs are two-dimensional retinal images that visualize the optic disc, macula, vasculature, and peripheral retina, while ultra-widefield retinal imaging extends the captured field to the far periphery in a single acquisition, surfacing lesions that conventional fields miss.

Optical coherence tomography (OCT) adds depth, producing cross-sectional B-scans and stacked volumetric cubes that resolve individual retinal layers at micron scale, and OCT angiography (OCT-A) reconstructs the retinal and choroidal microvasculature without dye injection. Fundus images are typically delivered as JPEG or PNG, or as DICOM, while OCT studies arrive as DICOM or vendor volume stacks with embedded acquisition metadata such as scan pattern, field of view, axial resolution, signal strength, laterality, and device model. Clinically valuable cohorts carry expert gradings: diabetic retinopathy severity on the International Clinical Diabetic Retinopathy (ICDR) scale, diabetic macular edema presence and severity, glaucoma indicators including cup-to-disc ratio and retinal nerve fiber layer (RNFL) thinning, age-related macular degeneration staging, retinopathy of prematurity, and hypertensive retinopathy.

The strongest datasets add pixel-level segmentation of vessels, the optic disc and cup, retinal layers on OCT, and lesions such as microaneurysms, hemorrhages, hard and soft exudates, and drusen, rather than image-level tags alone. Because retinal grading is subjective, robust datasets document inter-grader agreement, adjudication protocols, and label provenance, and balance disease prevalence so models do not overfit to majority-normal distributions. High-quality cohorts span diverse demographics, camera and OCT vendors, and image quality conditions, are quality-scored for gradability, and are rigorously de-identified to strip PHI from headers and any burned-in pixel text while preserving diagnostic fidelity.

On GetDATA, researchers and medtech companies post requests specifying modality (fundus, ultra-widefield, OCT, or OCT-A), field definition, laterality, label taxonomy and grading scale, annotation type (image-level, segmentation, or bounding box), class balance, and minimum case counts, and verified hospitals and labs fulfill them with compliant, quality-scored ophthalmology data. Established benchmarks and reference datasets include EyePACS, APTOS, Messidor-2, and IDRiD for diabetic retinopathy, alongside public OCT datasets for layer segmentation and macular pathology, which help align labels across institutions and support external validation. Whether you are building autonomous retinopathy screening, glaucoma risk stratification, or layer-segmentation pipelines, well-annotated multi-device cohorts are what separate models that generalize from those that fail silently.

Browse the open ophthalmology requests below, or explore related imaging categories.