NIH Scientists Revolutionize Eye Imaging with AI-Enhanced Ophthalmoscopy

NIH scientists use AI to sharpen standard eye imaging

National Institutes of HealthApr 23 2025 Scientists at the National Institutes of Health (NIH) have leveraged artificial intelligence to transform a device designed to see tissues in the back of the eye into one sharp enough to make out individual cells. The technique provides imaging resolution that rivals the most advanced devices available and is cheaper, faster, and doesn't require specialized equipment or expertise. The strategy has implications for early detection of disease and for the monitoring of treatment response by making what was once invisible now visible. "AI potentially puts next-generation imaging in the hands of standard eye clinics. It's like adding a high-resolution lens to a basic camera." said Johnny Tam, Ph.D., investigator at NIH's National Eye Institute and senior author of the study report, which published in Communications Medicine. Imaging devices, known as ophthalmoscopes, are widely used to examine the light-sensing retina in the back of the eye. A scanning laser ophthalmoscope is standard in eye clinics, but its resolution can only make out structures at the tissue level -- things such as lesions, blood vessels, and the optic nerve head. Next-generation ophthalmoscopes enabled with adaptive optics -- a technology that compensates for light distortion -- can make out cellular features, providing greater diagnostic information. However, adaptive optics-enabled imaging is still in the experimental phase. Tam and collaborators developed a custom AI system to digitally enhance images of a layer of tissue beneath the light-sensing photoreceptors, known as the retina's pigmented epithelium (RPE). The first step was to teach the system to recognize image quality as poor, moderate, or good. The researchers did this by feeding the system more than 1,400 images from different areas of the retina, obtained using adaptive-optics ophthalmoscopy. Next, they fed the system corresponding images from the same retinal locations but obtained using standard ophthalmoscopy. An image sharpness test showed that AI improved clarity eightfold. Our system used what it learned from rating the images obtained from adaptive optics to digitally enhance images obtained with standard ophthalmoscopy. It's important to point out that the system is not creating something from nothing. Features that we see in RPE cells with standard imaging are there, they're just unclear." Johnny Tam, Ph.D., investigator at NIH's National Eye Institute These techniques involve injection of a dye called indocyanine green (ICG) into the bloodstream to increase contrast of anatomical features. In the eye clinic, ICG is usually used to image the blood vessels of the eye. "Our ICG imaging strategy allows RPE cells to be quickly and routinely assessed in the clinic," said Joanne Li, Ph.D., first author of the report and a biomedical engineer in Tam's lab. "With AI, high quality images of the RPE cells can be obtained in a matter of seconds, using standard clinical imaging instruments." The RPE cells' function is to nourish and support photoreceptors. A variety of blinding conditions first affect RPE cells, including age-related macular degeneration, vitelliform macular dystrophy, and Stargardt disease. However, RPE cells cannot be easily imaged in the clinic. AI-enhanced ICG ophthalmoscopy puts RPE imaging within reach of the typical eye clinic.## National Institutes of Health Journal reference: Li, J., et al. (2025). Artificial intelligence assisted clinical fluorescence imaging achieves in vivo cellular resolution comparable to adaptive optics ophthalmoscopy. Communications Medicine. doi.org/10.1038/s43856-025-00803-z.

Supercharged ordinary clinical device gets a better look at the back of the eye

Scientists at the National Institutes of Health (NIH) have leveraged artificial intelligence to transform a device designed to see tissues in the back of the eye into one sharp enough to make out individual cells. The technique provides imaging resolution that rivals the most advanced devices available and is cheaper, faster, and doesn't require specialized equipment or expertise. The strategy has implications for early detection of disease and for the monitoring of treatment response by making what was once invisible now visible. "AI potentially puts next-generation imaging in the hands of standard eye clinics. It's like adding a high-resolution lens to a basic camera," said Johnny Tam, Ph.D., investigator at NIH's National Eye Institute and senior author of the study report, which was published in Communications Medicine. Imaging devices, known as ophthalmoscopes, are widely used to examine the light-sensing retina in the back of the eye. A scanning laser ophthalmoscope is standard in eye clinics, but its resolution can only make out structures at the tissue level -- things such as lesions, blood vessels, and the optic nerve head. Next-generation ophthalmoscopes enabled with adaptive optics -- a technology that compensates for light distortion -- can make out cellular features, providing greater diagnostic information. However, adaptive optics-enabled imaging is still in the experimental phase. Tam and collaborators developed a custom AI system to digitally enhance images of a layer of tissue beneath the light-sensing photoreceptors, known as the retina's pigmented epithelium (RPE). The first step was to teach the system to recognize image quality as poor, moderate, or good. The researchers did this by feeding the system more than 1,400 images from different areas of the retina, obtained using adaptive-optics ophthalmoscopy. Next, they fed the system corresponding images from the same retinal locations but obtained using standard ophthalmoscopy. An image sharpness test showed that AI improved clarity eightfold. "Our system used what it learned from rating the images obtained from adaptive optics to digitally enhance images obtained with standard ophthalmoscopy," said Tam. "It's important to point out that the system is not creating something from nothing. Features that we see in RPE cells with standard imaging are there, they're just unclear." These techniques involve injection of a dye called indocyanine green (ICG) into the bloodstream to increase contrast of anatomical features. In the eye clinic, ICG is usually used to image the blood vessels of the eye. "Our ICG imaging strategy allows RPE cells to be quickly and routinely assessed in the clinic," said Joanne Li, Ph.D., first author of the report and a biomedical engineer in Tam's lab. "With AI, high-quality images of the RPE cells can be obtained in a matter of seconds, using standard clinical imaging instruments." The RPE cells' function is to nourish and support photoreceptors. A variety of blinding conditions first affect RPE cells, including age-related macular degeneration, vitelliform macular dystrophy, and Stargardt disease. However, RPE cells cannot be easily imaged in the clinic. AI-enhanced ICG ophthalmoscopy puts RPE imaging within reach of the typical eye clinic.