AI-driven analysis of retinal images may eventually support earlier risk stratification for dementia and stroke, but current evidence remains insufficient to show that these tools improve clinical decisions and patient outcomes, according to a JAMA Ophthalmology Viewpoint.
The authors noted that nearly half of dementia cases may be preventable or delayed through modification of risk factors, and that up to 80% of strokes can be prevented with timely detection and risk-factor management. Carol Y. Cheung, PhD, and Vincent C.T. Mok, MD, of The Chinese University of Hong Kong, described AI-driven "oculomics" as a potential approach for assessing brain health through retinal imaging, because the retina shares embryologic origins, anatomical structures, and physiologic characteristics with the central nervous system and is the only part of it that can be visualized directly, noninvasively, and in vivo.
The authors cited prior studies suggesting that AI models using retinal photographs or optical coherence tomography images may help detect Alzheimer dementia and early cognitive impairment. They described similar potential in stroke risk assessment: in one cited study, an AI model detected silent brain infarctions from retinal photographs, and adding retinally inferred silent brain infarction status improved prediction of incident and recurrent stroke compared with clinical data alone. The authors noted further possible applications in Parkinson disease, AI-estimated "retinal biological age," and longitudinal monitoring of retina-brain changes, while noting that the evidence remains limited for some uses.
The sharper argument, though, is aimed at the field itself. The authors emphasized that most oculomics models remain in early clinical evaluation and have largely been judged by technical performance metrics such as area under the receiver operating characteristic curve, sensitivity, and specificity — measures they argued are insufficient on their own to establish clinical usefulness. Future efforts, they wrote, should move beyond classification to evaluate "whether oculomics can meaningfully improve clinical decision-making, downstream patient outcomes, and the overall benefit-harm balance of care in routine clinical practice." Both authors disclosed commercial ties to a company developing AI-derived retinal measures for brain disease evaluation.
They also called for clearer definitions of intended use, target population, and clinical setting before broader deployment. On the regulatory side, it means defining how such tools are classified and labeled — and what level of evidence, comparators, and endpoints should be required — depending on whether a tool is positioned for risk stratification, prescreening, triage, or monitoring.
The authors stressed that oculomics should complement, not replace, existing or emerging screening strategies such as neuroimaging, with possible roles in prescreening, triage, risk stratification, or longitudinal monitoring. They wrote that realizing the potential of oculomics will require further work in clinical validation, workflow integration, infrastructure development, and regulatory alignment.
Disclosures: Dr. Cheung reported being cofounder of i-Cognitio Sciences and having a US patent issued. Dr. Mok reported being founder of i-Cognitio Sciences, which develops AI-derived retinal measures for evaluating brain diseases.
Source: JAMA Ophthalmology
