The two most common retinal conditions that cause a progressive and often irreversible loss of central vision are age-related macular degeneration (AMD) and diabetic macular edema (DME). Fluid seeping into the macula is the most prevalent way that DME manifests as a consequence of diabetic retinopathy. Macular tissue wasting is a part of the aging process known as AMD. Both conditions significantly damage central vision, and in order to stop visual decline, early detection and long-term surveillance are necessary. The main non-invasive imaging method used by ophthalmologists to identify and track these conditions is optical coherence tomography (OCT), which provides high-resolution cross-sectional images of the retina that show microstructural alterations not visible with conventional imaging. OCT is not just required for initial diagnosis in the clinic, but also to monitor disease progression and treatment response. While public interest in using artificial intelligence (AI) to interpret OCT is growing, publicly available databases are often static, single-scan images without longitudinal structure and lacks clinical realism. This paper presents a high-resolution, longitudinal OCT dataset obtained directly from Alex iCare eye Center, a referral eye clinic, in a research partnership with Alamein International University. The dataset comprises 42,061 OCT scans in three diagnostic classes: 30,000 Normal eyes scans, 6,290 DME (Dry and Wet) scans, and 5,771 AMD scans. There are multiple scans in each patient folder arranged in visits folders, preserving the authentic history of retinal disease on real-world timelines. For consistency purposes and research utility, all the images were filtered, resized, and standardized. Scans were organized by diagnosis and patient ID. Low quality, duplicate images were manually removed. All data were anonymized and replaced by patient ID prior to processing for purposes of following research ethics and patient privacy regulations. This dataset includes a diverse range of machine learning applications like binary and multi-class classification, modeling temporal progression of diseases, predictive modeling, and decision support system building. Filling the gap between research studies and real-world clinical data, this dataset provides a valuable platform for advancing AI-driven diagnosis, prognosis, and tailored treatment in ophthalmology.
Paper
Full text
A Real-World OCT Imaging Dataset for AI-Based Classification of Retinal Diseases
Semantic Scholar · 2026
Abstract
The two most common retinal conditions that cause a progressive and often irreversible loss of central vision are age-related macular degeneration (AMD) and diabetic macular edema (DME). Fluid seeping into the macula is the most prevalent way that DME manifests as a consequence of diabetic retinopathy. Macular tissue wasting is a part of the aging process known as AMD. Both conditions significantly damage central vision, and in order to stop visual decline, early detection and long-term surveillance are necessary. The main non-invasive imaging method used by ophthalmologists to identify and track these conditions is optical coherence tomography (OCT), which provides high-resolution cross-sectional images of the retina that show microstructural alterations not visible with conventional imaging. OCT is not just required for initial diagnosis in the clinic, but also to monitor disease progression and treatment response. While public interest in using artificial intelligence (AI) to interpret OCT is growing, publicly available databases are often static, single-scan images without longitudinal structure and lacks clinical realism. This paper presents a high-resolution, longitudinal OCT dataset obtained directly from Alex iCare eye Center, a referral eye clinic, in a research partnership with Alamein International University. The dataset comprises 42,061 OCT scans in three diagnostic classes: 30,000 Normal eyes scans, 6,290 DME (Dry and Wet) scans, and 5,771 AMD scans. There are multiple scans in each patient folder arranged in visits folders, preserving the authentic history of retinal disease on real-world timelines. For consistency purposes and research utility, all the images were filtered, resized, and standardized. Scans were organized by diagnosis and patient ID. Low quality, duplicate images were manually removed. All data were anonymized and replaced by patient ID prior to processing for purposes of following research ethics and patient privacy regulations. This dataset includes a diverse range of machine learning applications like binary and multi-class classification, modeling temporal progression of diseases, predictive modeling, and decision support system building. Filling the gap between research studies and real-world clinical data, this dataset provides a valuable platform for advancing AI-driven diagnosis, prognosis, and tailored treatment in ophthalmology.