Objective:

To investigate similarities in retinal aging patterns between humans and non-human primates using deep learning-based age prediction and attention-based saliency analysis. 

Background:

Non-human primates are the premier animal model for studying human aging and retinal biology due to their genetic and metabolic similarities, the presence of a true macula, and shared age-related pathologies. The retina provides a unique window into biological aging, with deep learning-based age predictions from retinal fundus images linked to systemic health outcomes in humans. Establishing cross-species validity of retinal aging biomarkers could provide a foundation for testing anti-aging interventions in primate animal models. 

Design/Methods:

We analyzed 2,210 retinal images from 180 rhesus macaques (filtered from 8971 images using AutoMorph) and age-matched human images from the Collaborative University of California Teleophthalmology Initiative, based on a 3:1 age conversion ratio. A DINOv3 vision transformer was fine-tuned using 10-fold cross-validation separately on each dataset. Attention rollout generated saliency maps were anatomically aligned using optic disc and foveal coordinates derived from VascX and then averaged. We compared anatomical region-based saliency distribution between the two species, then performed cross-species model validation by training on one species and testing on the other. 

Results:

Our model age predictions achieved a mean absolute error (MAE) of 2.2 years (R²=0.73) for macaques and MAE=3.62 years (R²=0.86) for humans. Averaged saliency maps showed qualitatively similar attention patterns between the two species, highlighting the peripapillary region and the retinal vasculature. Quantitative analysis showed greater attention to the optic disc and vasculature in macaques and slightly lower macular attention compared to humans. Cross-species model evaluation demonstrated a correlation of r=0.741 when training on primates and testing on humans, and r=0.597 for the reverse direction. 

Conclusions:

Deep learning models reveal similar retinal aging patterns between humans and macaques, supporting the use of nonhuman primates as translational models for aging research and therapeutic development.