A Heart study has shown that “retinal vascular fingerprints” could be used as a non-invasive biomarker to predict incident stroke risk in patients as accurately as other traditional risk stratification models requiring more invasive lab tests.
To examine the associations between incident stroke and retinal vascular parameters extracted from the UK Biobank’s catalogue of 68,753 fundus photographs, the multi-institutional team of researchers developed RMHAS (Retina-based Microvascular Health Assessment System), a software system that integrates deep learning to analyze retinal images at high volume.
The “fingerprint” identified consisted of 29 novel retinal indicators, all of which could be significantly linked to first-time stroke risk, and could “offer new avenues for stroke pathophysiology research,” the authors say.
RMHAS “only requires an age, sex, and retinal scan, which offers a non-invasive and easily implementable screening method by leveraging existing ophthalmology infrastructure and the regular nature of eye check-ups,” explains study author Danli Shi, of The Hong Kong Polytechnic University.
Shi points out that, while stroke affects over 100 million people globally, nearly 90 percent of these occurrences can be attributed to modifiable risk factors – such as high cholesterol, diet, hypertension, and smoking – meaning that they could, in theory, be preventable.
“We hope the findings from this study could catalyse a new paradigm of interdisciplinary collaboration between ophthalmologists and neurologists,” Shi goes on. “Rather than working in separate silos, these specialists could develop integrated care pathways, especially incorporating stroke screening into retinal screening programs where patients with concerning retinal vascular patterns could be referred to neurologists for further evaluation and early intervention.”
This method could prove especially valuable in low-resource and primary healthcare settings, potentially replacing traditional stroke risk assessments that require clinicians to take multiple clinical measurements and blood tests. “Deep learning models like RMHAS represent the vanguard of a revolution in medical imaging analysis,” adds Shi. “In the future, these systems could evolve into real-time diagnostic tools that provide instant analysis during routine eye examinations. They could be integrated into standard ophthalmological equipment, offering immediate risk assessments and suggesting referral pathways based on detected abnormalities.”
