Researchers have developed a deep learning biomarker that demonstrates robust associations with individual mortality risk by analyzing low-cost retinal photographs. The AI model, called RetiME (Retinal Mortality Equivalent), represents a research advance in aging assessment that could potentially be integrated into routine eye exams worldwide pending further evaluation.

From Eye Scans to Mortality Forecasts

RetiME leverages large foundation models to map retinal features directly to mortality risk from non-invasive retinal photographs. The system goes beyond simple risk scoring by reconstructing personalized lifespan trajectories using a Gompertz-Makeham mortality modeling framework.

The model was trained on European cohorts and validated across four independent populations spanning multiple continents: the UK Biobank (European), SEED study (Southeast Asia), UEMS (Eurasia), and AREDS (North America). This multi-ethnic validation demonstrates the technology's potential generalizability across diverse healthcare systems and populations.

According to the research published in Nature Biomedical Engineering, RetiME showed robust associations with all-cause mortality, cause-specific mortality, and incidence of systemic and ocular diseases. The training data was accessed under UK Biobank application numbers 45925 and 86091, while AREDS data is publicly available through the Database of Genotypes and Phenotypes (dbGaP).

Three Distinct Aging Patterns Identified

One of RetiME's most significant findings emerged from trajectory clustering analysis, which revealed three distinct mortality patterns: cancer-predominant, cardiometabolic, and neurodegenerative risks. This classification suggests that individuals age along different biological pathways, each characterized by specific disease vulnerabilities.

The researchers also conducted genome-wide analyses that identified genetic loci associated with RetiME scores, potentially linking retinal features to underlying genetic determinants of aging and mortality risk.

A Practical Alternative to Molecular Biomarkers

This technology offers a potentially scalable, affordable screening tool for aging populations using existing retinal imaging infrastructure already present in ophthalmology and primary care settings. Unlike expensive molecular biomarkers such as DNA methylation clocks, retinal imaging represents a non-invasive, widely accessible approach to mortality risk assessment.

The findings support further evaluation of retinal imaging and trajectory modeling as one component of comprehensive mortality risk assessment. Healthcare providers could potentially integrate RetiME into routine eye exams to stratify patients for preventive interventions, while health tech companies may explore commercial applications for population health management and risk stratification, pending additional validation.

The code for RetiME has been made available via GitHub at https://github.com/xt0402/RetiME, potentially enabling broader research and clinical evaluation. The convergence of affordable imaging technology, deep learning capabilities, and existing clinical infrastructure positions RetiME as a research tool warranting further evaluation for cost-effective mortality risk assessment across diverse global populations.

Sources

  1. Re-timing mortality risk through individualized modelling of lifespan trajectories — Nature Biomedical Engineering