There may be some truth to the saying "the eyes are the window to the soul." Age-related changes are reflected in the retina, the light-sensitive tissue at the back of the eye. Recent research shows that a photo of the retina may also reveal potential risks for major diseases like diabetes. A research group led by Professor Toru Nakazawa at the Tohoku University Graduate School of Medicine has developed an artificial intelligence (AI) model that estimates "retinal age," an indicator reflecting an individual's biological aging, from a single fundus photograph of the retina. The apparent link between retinal age and major disease risk could be used as a helpful screening aid in the future.
"Fundus images are non-invasive photos of the eye taken as part of regular health check-ups - so no additional work is needed," explains Nakazawa, "Our model would be a nearly frictionless addition to a clinician's typical workflow."
The AI model was trained on 50,595 quality-controlled fundus images from disease-free adults and internally validated on 7,288 additional images. The AI model assesses "retinal age" based on features seen from these fundus images. The researchers found it was more accurate than previous benchmarks at guessing the age of patients, with an average error of around three years. The blood sugar marker HbA1c was incorporated during training to help capture age-related retinal patterns more robustly. However, no blood test would be needed if the model was deployed for clinical use. It would be as simple as taking a photo.
The researchers then examined the "retinal age gap," defined as the difference between AI-predicted retinal age and chronological age. While the model was very good at accurately predicting actual age based on the retina, they noticed there was a larger gap for certain patients. After matching participants by age and sex, they found that this gap was significantly larger in individuals with diabetes, heart disease, or a history of stroke, indicating a tendency for the retina to appear older than expected for their chronological age.
The researchers note that while these findings are exciting, they are primarily based on cross-sectional analyses, suggesting correlation but not causation. Further prospective longitudinal studies will be needed to determine the extent to which retinal age and the retinal age gap can predict future disease onset.
"We are already planning a study that follows a cohort of over 10,000 individuals with continuous 3-year follow-up to examine whether retinal age-related signals are associated with the future development of cardiovascular and other systemic diseases." says Nakazawa.
The tool is designed to one day serve as a promising screening aid, revealing patients who may need further health assessments or other personalized prevention strategies, depending on the clinician's discretion.
External validation of the study was conducted with support from the lab of Professor Pearse Keane at University College London, a strategic partner of Tohoku University. The findings were published in Communications Medicine on April 8, 2026.
- Publication Details:
Title: High-accuracy retinal age prediction via fundus-based multitask learning reveals the effect of systemic disease
Authors: Takahiro Ninomiya, Akiko Hanyuda, Naoki Kiyota, Parmanand Sharma, Yukun Zhou, Siegfried K. Wagner, Keita Suzuki, Takanari Nozaki, Takehiro Miya, Naoki Takahashi, Kazuko Omodaka, Noriko Himori, Yoichi Ichikawa, Pearse A. Keane, Toru Nakazawa
Journal: Communications Medicine
